CIGRÉ Technical Brochure

Advanced Angle Stability Controls

Prepared byTask Force 17

of Advisory Group 02of Study Committee 38

December 1999

International Conference on Large High Voltage Electric SystemsConférence Internationale des Grands Réseaux Électriques a Haute Tension

CIGRÉ TF 38.02.17

Advanced Angle Stability ControlsConvenor: Carson Taylor (USA)

Members and contributors

Florencio Aboytes (Mexico) Magnus Akke (Sweden)Tom Anderson(Scotland) Göran Andersson (Sweden)Miroslav Begovoc (USA) Bharat Bhargava (USA)Henri Bourles (France) Fabio Casamatta (Italy)Joe Chow (USA) Sandro Corsi (Italy)Jeff Dagle (USA) Art DeGroff (USA)Peter Donalek (USA) Carlos Gama (Brazil)Suma Geeves (UK) Jan Ove Gjerde (Norway)Paulo Gomes (Brazil) Michael Hadingham (South Africa)Adel Hammad (Switzerland) John Hauer (USA)Michael Henderson (USA) David Hill (Australia)Takashi Hiyama (Japan) Michael Hughes (UK)Satoru Ihara (USA) Lawrence Jones (Sweden)Nelson. Zeni Junior (Brazil) Innocent Kamwa (Canada)Daniel Karlsson (Sweden) Prabha Kundur (Canada)Kwang Lee (USA) Donald Macdonald (UK)Thibault Margotin (France) Nelson Martins (Brazil)Jim McCalley (USA) Dagmar Niebur (USA)Mojtaba Noroozian (Sweden) Damir Novosel (USA)Torben Østrup (Denmark) John Paserba (USA)Paulo Paiva (Brazil) Mania Pavella (Belgium)Mrinal Pal (USA) Steve Rovnyak (USA)Juan Sanchez-Gasca (USA) Pierangelo Scarpellini (Italy)Dick Schulz (USA) Jim Smith (USA)Hisao Taoka (Japan) Kjetil Uhlen (Norway)Ebrahim Vaahedi (USA) Lei Wang (Canada)Louis Wehenkel (Belgium)

Contents

1. Introduction and Survey1.1 Review of power system synchronous stability basics.............................1-21.2 Concepts of power system stability controls ............................................1-61.3 Types of power system stability control and possibilities

for advanced control .................................................................................1-91.4 Dynamic security assessment .................................................................1-181.5 Intelligent controls..................................................................................1-191.6 Effects of industry restructuring on stability controls ............................1-191.7 Experience from recent power failures...................................................1-201.8 Coordination with other CIGRÉ and industry work...............................1-201.9 Summary ................................................................................................1-21

2. Advanced Linear and Nonlinear Control Design2.1 Nonlinear control......................................................................................2-22.2 Linear control techniques .........................................................................2-4

3. State-of-the-Art in Digital Control3.1 Review of digital control of dynamic systems .........................................3-33.2 Basic structure of digital control systems.................................................3-83.3 Evolution of excitation control systems through

microprocessor technology.....................................................................3-113.4 Application of digital control to SVCs...................................................3-143.5 Trends in digital control .........................................................................3-19

4. State-of-the-Art in Intelligent Controls4.1 Fuzzy systems for power system control..................................................4-24.2 ANN for power system control ................................................................4-64.3 Decision trees for power system control ................................................4-16

5. Integration of Dynamic Security Assessment and Stability Controls5.1 On-line dynamic stability assessment design ...........................................5-15.2 Other integration of DSA and stability controls.......................................5-9

6. Measurement and Communication Technology6.1 Introduction to transducers .......................................................................6-36.2 The signal environment for power system transducers ............................6-66.3 Signal processing in power system transducers .....................................6-106.4 Criteria and procedures for evaluating transducer performance.............6-136.5 Transducer modeling and simulation .....................................................6-146.6 Digital transducers and phasor measurements .......................................6-15

6.7 The transducers as an intelligent electronic device ................................6-176.8 Role of communication channels in wide-area control ..........................6-176.9 Observed performance of digital communications in the BPA

phasor measurement network.................................................................6-196.10 Future digital communication for stability control.................................6-226.11 Optical sensors .......................................................................................6-22

7. Applications of Advanced Controls7.1 Brazilian north–south interconnection—application of thyristor

controlled series compensation (TCSC) to damp interareaoscillation mode .......................................................................................7-1

7.2 Analysis and control of Yimin–Fengtun 500-kV TCSC system ..............7-37.3 Wide-area stability control .......................................................................7-47.4 Active load modulation for stability control ............................................7-77.5 Active power modulation of generators and energy storage

for oscillatory instability control ..............................................................7-8

8. Stability Controls with Industry Restructuring8.1 Some examples of new scenarios.............................................................8-18.2 Coordinated planning and operation in a competitive environment ........8-98.3 The impact of IPP thermal generation on system dynamic

performance............................................................................................8-108.4 Other issues related to power system performance in the new

utility environment .................................................................................8-14

9. Conclusions9.1 Conclusions ..............................................................................................9-19.2 Areas for future work ...............................................................................9-2

AppendicesA Adjustable speed hydro generation..........................................................A-1B Dynamic performance changes produced by numerical integration

algorithms................................................................................................ B-1C Space vector, positive and negative sequence vectors ............................ C-1D Sideband production in RMS calculations ..............................................D-1E Basic phasor calculations ........................................................................ E-1F Laboratory evaluations of power system transducers...............................F-1G Field evaluations of power system transducers .......................................G-1H Transducer modeling and simulations.....................................................H-1I Performance of BPA analog communication channels .............................I-1J A new look at damping control ................................................................ J-1

Chief Editor C. W. Taylor

Assistant Editors/Readers G. Andersson and A. E. Hammad

Chapter contributorsChapter 1: C. W. Taylor (lead editor), H. Taoka, A. E. Hammad, R. P.

Schulz, A. G. DeGroff

Chapter 2: J. R. Smith (lead editor), M. Akke, L. E. Jones

Chapter 3: M. Noroozian (lead editor), S. Corsi, P. Paiva

Chapter 4: D. Novosel (lead editor), D. Niebur, S. M. Rovnyak, J. D.McCalley, T. Hiyama, K. Y. Lee, S. Ihara, C. W. Taylor

Chapter 5: E. Vaahedi (lead editor), P. Kundur, L. Wang, M. Pavella,P. Scarpellini, K. Cheung, C. W. Taylor

Chapter 6: J. F. Hauer (lead editor), C. W. Taylor

Chapter 7: M. Akke (lead editor), L. E. Jones, T. Østrup, I. Kamwa, C.Gama, B. Bhargava, M. K. Pal, R. P. Schulz, C. W. Taylor

Chapter 8: P. Gomes (lead editor), F. P. de Mello, N. Martins, X. Vieira F.,J. F. Hauer, T. Østrup, M. Henderson, K. Uhlen, C. W. Taylor

Chapter 9: C. W. Taylor (lead editor)

Appendix A P. J. Donalek

Appendix B S. Corsi

Appendix C M. Noroozian

Appendices D–I J. F. Hauer

Appendix J M. K. Pal

Chapter 1

Introduction and Survey

Power system synchronous or angle instability phenomenon limits power transfer,especially where transmission distances are long. This is well recognized and manymethods have been developed to improve stability and increase allowable power transfers[1-1,1-2,1-3]. Section 1.1 reviews the basics of power system stability.

The synchronous stability problem has been fairly well solved by fast fault clearing,thyristor exciters, power system stabilizers, and a variety of other stability controls suchas generator tripping. Fault clearing of severe short circuits can be less than three cycles(50 ms for 60 Hz frequency). The effect of the faulted line outage on generatoracceleration and stability may be greater than that of the short circuit itself.

Nevertheless, requirements for more intensive use of available generation andtransmission, more onerous load characteristics, greater variation in power schedules, andindustry restructuring pose new challenges. Recent large-scale power failures in Northand South America and in other parts of the world have heightened the concerns.

This report on advanced angle stability controls provides industry guidance in solvingstability problems with new or relatively new technologies. The technologies includecontrol theory and applications, power electronics, microprocessor controllers, signalprocessing, digital and optical transducers, and telecommunications. There is greatopportunity for synergism in these areas. The goals are new control strategies that areeffective and robust. Effective in an engineering sense means “cost-effective.” Controlrobustness is the capability to function appropriately for a wide range of power systemoperating and disturbance conditions.

Much can be gained by technology transfer to the electric power industry from disciplinessuch as automatic control, artificial intelligence, and signal processing.

Power system engineers responsible for determining stability-related transfer limits andfor developing means for extending transfer limits are always acquainted with state-of-the-art control technology. Protection or other engineers responsible for implementationof stability controls may not be entirely familiar with control technology or power systemstability phenomena. This report provides guidance on advanced methods to improvestability.

The initial incentives for this report were advances in synchronized voltage phase anglemeasurements and in high voltage power electronic equipment to directly or indirectlycontrol transmission voltage and generator rotor angles. These concepts were discussed ata panel session on “More Effective Networks” at the 1996 general meeting in Paris; thepanel session involved eight study committees. Christensen further described suchconcepts in [1-4]. An interesting question arose:

• What is the value of direct control of voltage phase angle? Equipment such as power-electronic controlled series compensation and phase-shifting transformers maydirectly control the phase angle (and indirectly control generator rotor angles).

1-2

A more comprehensive review of advanced technology for stability control is, however,desirable. Our emphasis in this report is on angle stability, but there is a close relationbetween voltage magnitude control and angle stability. Our emphasis is also on largedisturbances and nonlinear aspects of stability control. The techniques described areapplicable to practical large-scale power systems.

This introductory chapter surveys the field of power system stability controls, and thepossibilities for advanced angle stability controls that are described in the followingchapters.

1.1 Review of Power System Synchronous Stability BasicsMany publications, for example references 1-1, 1-2, and 1-5, describe the basics—whichwe briefly review here. Power generation is largely obtained by synchronous generators,which may be interconnected over thousands of kilometers in very large power systems.All generators must operate in synchronism during normal and disturbance conditions.Loss of synchronism of a generator or a group of generators with respect to another groupof generators is instability that could result in expensive widespread power blackouts.

The essence of synchronous stability is balance of individual generator electrical andmechanical torques as described by Newton’s second law applied to rotation:

em TTdt

dJ −=ω

,

where J is moment of inertia of the generator and prime mover, ω is speed, mT is

mechanical prime mover torque, and eT is electrical torque related to generator electric

power output. The generator speed determines the generator rotor angle changes relativeto other generators. Figure 1-1 shows the basic “swing equation” block diagramrelationship for a generator connected to a power system.

Tm Tacc

H2

1

Te

α

GeneratorElectricalEquations

PowerSystem

Disturbances

δo

∆ω ∫ • dt oω∫• dt

+

Fig. 1-1. Block diagram of generator electromechanical dynamics.

δ

The block diagram representing the internal generator dynamics is explained as follows:

1-3

• The inertia constant, H, is proportional to the moment of inertia and is the kineticenergy at rated speed divided by the generator MVA rating. Units are MW-seconds/MVA (or seconds).

• mT is mechanical torque in per unit. As a first approximation it’s assumed to beconstant. It is, however, influenced by speed controls (governors) and prime moverand energy supply system dynamics.

• oω is rated frequency in radians/second (2πfo, where fo is rated frequency in Hz).

• oδ is pre-disturbance rotor angle in radians relative to a reference generator.

• The power system block comprises the transmission network, loads, power electronicdevices, and other generators/prime movers/energy supply systems with their controls.The transmission network is generally represented by algebraic equations. Loads andgenerators are represented by algebraic and differential equations.

• Disturbances include short circuits, and line and generator outages. A severedisturbance is a three-phase short circuit near the generator. This causes electricpower and torque to be zero, with accelerating torque Tacc equal to mT . (Althoughgenerator current is very high during the short circuit, its power factor, and activecurrent and active power are close to zero.)

For illustration, a simple conceptual transmission model as shown in Fig. 1-2 is used. Itcomprises a remote generator connected to a large power system by two paralleltransmission lines with an intermediate switching station. With some approximationsadequate for a time of one second or more following a disturbance, the Figure 1-3 blockdiagram is realized. The basic relationship between power and torque is ωTP = . Sincespeed changes are quite small, power is considered equal to torque in per unit. Thegenerator representation is a constant voltage, E ′ , behind a reactance. The transformerand transmission lines are represented by inductive reactances. Using the

relation *IES ′= , the generator electrical power has the well-known relation:

δsine X

VEP

′= ,

where V is the large system (infinite bus) voltage and X is the total reactance from thegenerator internal voltage to the infinite bus. The above equation approximatescharacteristics of a detailed, large-scale model, and illustrates that the power system isfundamentally a highly nonlinear system for large disturbances.

Figure 1-4 shows the relation between Pe and δ graphically. The pre-disturbanceoperating point is at the intersection of the load or mechanical power characteristic andthe electrical power characteristic. Normal stable operation is at oδ . For example, a small

increase in mechanical power input causes an accelerating power ( em PP − ) that increases

δ to increase eP until accelerating power returns to zero at a slightly different

1-4

equilibrium point. The opposite is true for the unstable operating point at oδπ − : a small

increase in mechanical power will cause a runaway increase in angle.

The angle oδ is generally less than 45°. For small disturbances, the above power-angle

equation can be linearized ( δδ ≅sin in radians for angles under 30°). The block diagram(Figure 1-3) would then represent a second order differential equation oscillator. For aremote generator connected to a large system the oscillation frequency is 0.8–1.1 Hz.

+

+

~

Fig. 1-2. Remote power plant to large system. Short circuit location is shown.

Pm

-Pe

δo

∆ω ∫ • dtoω

+

Fig. 1-3. Simplified block diagram of generator electromechanical dynamics.

δ ∫ • dt

H2

1

)sin(•′X

VE

De

E’ ∠ δ V ∠ 0

-

Dm

During normal operation, mechanical and electrical torques are equal and the generatorruns at a constant frequency close to 50 or 60 Hz rated frequency. If, however, a shortcircuit occurs on a transmission line the electric power output will be momentarilypartially blocked from reaching loads and the generator (or group of generators) willaccelerate, with increase in generator speed and angle. If the acceleration relative to othergenerators is too large, synchronism will be lost. Loss of synchronism is an unstable,runaway situation with large variations of voltages and currents that will normally causeprotective separation of a generator or a group of generators. Following clearing of theshort circuit by line removal, the increase in the electrical torque (and power) developedas the angle increases will decelerate the generator. If deceleration reverses the angleswing prior to oδπ ′− , stability can be maintained at a new operating point oδ ′ (Figure 1-

4). If the angle swing is beyond oδπ ′− , accelerating power/torque again becomes

positive resulting in a runaway increase of angle and speed, and thus instability.

1-5

Figure 1-4 illustrates the equal area stability criterion for “first swing” stability. If thedecelerating area (energy) above the mechanical power load line is greater theaccelerating area below the load line, stability can be maintained.

Stability controls help maintain stability by decreasing the accelerating area or increasingthe decelerating area. This may be achieved during the forward angle swing by increasingthe electrical power output, or by decreasing the mechanical power input, or by both.

δ

∆ω

πδ

Fig. 1-4. (a) Power angle curve and equal area criterion. Dark shading for accelerationenergy during fault. Light shading for additional acceleration energy because of lineoutage. Black shading for deceleration energy. (b) Angle–speed phase plane. Dottedtrajectory is for unstable case.

Pm

P

Fault on electricalpower

Post-disturbanceelectrical power

Pre-disturbanceelectrical power

oδ oδ ′

oδ

(a)

(b)

oδ ′

Unstable

Stable

Figure 1-3 also shows mechanical and electrical damping paths (dashed, damping powerin phase with speed) that represent oscillation damping mechanisms respectively in theprime-mover and generator, loads, and other devices. For positive ∆ω the mechanical

1-6

damping, including friction and windage losses, reduces the mechanical input torquewhereas the electrical damping enhances the electrical output torque. Controls, notablygenerator automatic voltage regulators with high gains, can introduce negative electricaldamping at some oscillation frequencies. (In any feedback control system, high gaincombined with time delays can cause positive feedback and instability.) For stability, thenet damping must be positive for both normal conditions and for large disturbances withoutages.

External stability controls may also be added to improve damping.

The above analysis can be generalized to large interconnected systems. For first swingstability, synchronous stability between two critical groups of generators is usually ofconcern. For damping, many oscillation modes are present, all of which require positivedamping. The low frequency modes (0.1–0.8 Hz) associated with interarea oscillationsbetween large portions of a power system are the most difficult to damp.

1.2 Concepts of Power System Stability ControlsFigure 1-5 shows a general structure for analysis of power system stability, and fordevelopment of power system stability controls.

-

+

response detection

u

∆ y

structural changes

structural changes∆ Pdirect

detection

switched load

switched gen

Power SystemDisturbances

FeedforwardControls

PowerSystem

Dynamics

∆ load

∆ gen

FeedbackControls

Fig. 1-5. General power system structure showing stability controls [1-8].

SystemVariables

Stability problems typically involve disturbances such as short circuits, with subsequentremoval of faulted elements. Generation or load may be lost, resulting in generation–loadimbalance and frequency excursions. These disturbances stimulate power systemelectromechanical dynamics. Improperly designed or tuned controls may contribute to

1-7

stability problems; as mentioned, one example is negative damping torques caused bygenerator automatic voltage regulators.

Because of power system synchronizing and damping forces (including the feedbackcontrols shown on Figure 1-5), stability is maintained for most disturbances and operatingconditions.

Feedback controls. The most important feedback (closed-loop) controls are the generatorexcitation controls (automatic voltage regulator often including power system stabilizer).Other feedback controls include prime mover controls, controls for reactive powercompensation such as static var systems, and special controls for HVDC links. Thesecontrols are usually linear, continuously active, and based on local measurements.

There are, however, interesting possibilities for very effective discontinuous feedbackcontrols, with microprocessors facilitating implementation. Discontinuous controls havecertain advantages over continuous controls. Continuous feedback controls are potentiallyunstable. In complex power systems, continuously-controlled equipment may causeadverse modal interactions [1-7,8]. Modern digital controls, however, can bediscontinuous, and take no action until certain monitored variables are out-of-range. Thisis analogous to the very effective biological systems that operate on the basis of excitatorystimuli [1-9].

Bang-bang discontinuous control can operate several times to control large amplitudeoscillations, providing time for linear continuous controls to become effective.

Feedforward controls. Also shown on Figure 1-5 are specialized feedforward (open-loop) controls that are a powerful stabilizing force for severe disturbances and for highlystressed operating conditions. Short circuit or outage events can be directly detected toinitiate pre-planned actions such as generator or load tripping, or reactive powercompensation switching. These controls are rule-based, with rules developed fromsimulations (i.e., pattern recognition). These “event-based” controls are very effectivesince rapid control action prevents electromechanical dynamics from becoming stabilitythreatening.

“Response-based” feedforward controls are also possible. These controls initiatestabilizing actions for arbitrary disturbances that cause significant “swing” of measuredvariables.

Feedforward controls such as generator or load tripping can ensure a post-disturbanceequilibrium with sufficient region of attraction. With fast control action the region ofattraction can be small compared to requirements with only feedback controls.

Feedforward controls have been termed discrete supplementary controls [1-5], specialstability controls [1-3], special protection systems [1-10], remedial action schemes, andemergency control systems [1-11].

Generally speaking, feedforward controls can be very powerful. Although the reliabilityof special stability controls is often an issue [1-12], adequate reliability can be obtainedby careful design. Controls are typically required to be as reliable as primary protectiverelaying. Duplicated or multiple sensors, redundant communications, and duplicated or

1-8

voting logic are common [1-13]. Response-based controls are often less expensive thanevent-based controls because fewer sensors and communication paths are needed.

Undesired operation by some feedforward controls are relatively benign, and controls canbe “trigger happy.” For example, infrequent misoperation or unnecessary operation ofHVDC fast power change, reactive power compensation switching, temporary fastvalving of fossil units, and transient excitation boosting may not be very disruptive.Misoperation of generator tripping (especially of steam-turbine generators), fast valving,load tripping, or controlled separation, however, are disruptive and costly.

Synchronizing and damping torques. Power system electromechanical stability meansthat synchronous generators and motors must remain in synchronism followingdisturbances — with positive damping of rotor angle oscillations (“swings”). For verysevere disturbances and operating conditions, loss of synchronism (instability) occurs onthe first swing within about one second. For less severe disturbances and operatingconditions, instability may occur on the second or subsequent swings because of acombination of insufficient synchronizing and damping torques at synchronous machines.

Effectiveness and robustness. Power systems have many electromechanical oscillationmodes, and each mode can potentially become unstable. Lower frequency interarea modesare the most difficult to stabilize. Controls must be designed to be effective for one ormore modes and must not cause adverse interaction for other modes.

There are recent advances in robust control theory, especially for linear systems. For realnonlinear systems, emphasis should be on knowing uncertainty bounds and on sensitivityanalysis using detailed nonlinear, large-scale simulation. For example, the sensitivity ofcontrols to different operating conditions and load characteristics should be studied. On-line simulation using actual operating conditions reduces uncertainty, and can be used forcontrol adaptation.

Actuators. Actuators may be mechanical or power electronic. There are tradeoffsbetween cost and performance. Mechanical actuators (circuit breakers) are lower cost, andare usually sufficiently fast for electromechanical stability (e.g., two-cycle opening time,five-cycle closing time). They have restricted operating frequency and are generally usedfor feedforward controls.

Circuit breaker technology and reliability have improved in recent years [1-14,1-15].Bang-bang control (up to perhaps five operations) for interarea oscillations with periodsof two seconds or longer is feasible [1-16]. Mechanical switching has traditionally usedsimple relays, but with advanced technologies and intelligent controls [1-17], it canapproach or even exceed the sophistication of controls of, for example, thyristor-switchedcapacitor banks.

Power electronic phase control or switching using thyristors has been widely used ingenerator exciters, HVDC links, and static var compensators. Newer devices, especiallygate-turnoff thyristors, now have voltage and current ratings sufficient for high powertransmission applications (other semiconductor devices with current turnoff capabilitiesare available at lower power ratings). Advantages of power electronic actuators are veryfast control, unrestricted switching frequency, and minimal transients and maintenance.

1-9

For economy, existing actuators, perhaps supplemented with intelligent controls, shouldbe used to the extent possible. These include generator excitation and prime moverequipment, HVDC transmission equipment, and circuit breakers. For example, infrequentgenerator tripping may be cost-effective compared to new power electronic actuatedequipment.

Reliability criteria. Experience shows that instability incidents are usually not caused bythree-phase faults near large generating plants that are typically specified in deterministicreliability criteria. Rather they are the result of a combination of unusual failures andunforeseen circumstances. The three-phase fault reliability criterion is often considered anumbrella criterion providing a sufficient stability margin for less predictable disturbancesinvolving multiple failures such as single-phase short circuits with “sympathetic” trippingof unfaulted lines. Of main concern is multiple related (common-mode) failuresinvolving lines on the same right-of-way or with common terminations.

Reliability criteria also provide a performance margin to account for the manyuncertainties in simulation analysis. Uncertainties can include modeling and data errors,and differences between the simulated and the actual operating conditions. Simulationsare usually off line, and are often performed several months before actual operation. On-line, near real-time simulations reduces operating condition uncertainty.

Reliability criteria margins can be, for example, a power margin on allowable transfer(typically 5%), or a voltage dip of no more than 20–30% during swings.

Purpose of stability controls. The purpose of stability controls is to remove stability as alimit on power transfers. Excessive investment to obtain high performance such as rapiddamping of oscillations is not desirable.

1.3 Types of Power System Stability Controls and Possibilities forAdvanced Controls

Stability controls are of many types including:

• Generator excitation controls

• Prime mover controls including fast valving

• Generator tripping

• Fast fault clearing

• High speed reclosing, and single-pole switching

• Dynamic braking

• Load tripping and modulation

• Reactive power compensation switching or modulation (series and shunt)

• Current and voltage injections by voltage source inverter devices (STATCOM, UPFC,SMES, battery storage)

1-10

• Fast voltage phase angle control

• HVDC link supplementary controls

• Adjustable-speed (doubly-fed) generation

• Controlled separation and underfrequency load shedding

We will summarize these controls. Chapter 17 of reference 1-2 provides considerableadditional information. Reference 1-18 describes use of many of these controls in Japan.

Excitation control. Generator excitation controls are a basic stability control. Thyristorexciters with high ceiling voltage provide powerful and economical means to ensurestability for large disturbances. Modern automatic voltage regulators and power systemstabilizers are digital, facilitating additional capabilities such as adaptive control andspecial logic [1-19–22].

Excitation control is usually based on local measurements. Therefore full effectivenessmay not be obtained for interarea stability problems where local measurements are notsufficient. Line drop compensation [1-23–24] is one method to increase the effectiveness(sensitivity) of excitation control, and to improve coordination with static varcompensators that normally control transmission voltage with small droops.

Several forms of discontinuous control have been applied to keep excitation field voltagenear ceiling levels during the first forward interarea swing [1-2,1-25,1-26]. Recalling theproposed use of angle measurement for stability control, the control described inreferences 1-2 and 1-25 computes change in rotor angle locally from the power systemstabilizer (PSS) speed change signal. The control described in reference 1-26 is afeedforward control that injects a decaying pulse into the voltage regulators at a largepower plant following direct detection of a large disturbance. Figure 1-6 showssimulation results using this Transient Excitation Boosting TEB.

Prime mover control including fast valving. Fast mechanical power reduction (fastvalving) at generators is an effective means of stability improvement. Use has beenlimited, however, because of the coordination required between characteristics of theelectrical power system, the prime mover and prime mover controls, and the energysupply system (boiler).

Digital prime mover controls facilitate addition of special features for stabilityenhancement. Digital boiler controls, often retrofitted on existing equipment, mayimprove the feasibility of fast valving. Although not common, turbine power can bemodulated by prime movers controls to improve damping of interarea oscillations.

Fast valving has been found to be lower cost than tripping of turbo-generators. References1-2 and 1-27 describe investigations and recent implementations of fast valving. In theAEP application at Rockport [1-27], temporary fast valving has been found to beattractive, since both the first cost and operating costs of these fast valving schemes areless than the best alternative, which include additional transmission circuits. AEP andseveral other utilities make continual use of this means of improving rotor angle stability,

1-11

50

100

150

200

250

0 2 4 6 8 10

Rel

ativ

e an

gle

- de

gree

s

Time - seconds

w/o TEB

w/ TEB

Fig. 1-6. Rotor angle swing of Grand Coulee Unit 19 in Pacific Northwest relative to theSan Onofre nuclear plant in Southern California. The effect of transient excitationboosting (TEB) at the Grand Coulee Third Power Plant following bipolar outage of thePacific HVDC Intertie (3100 MW) is shown [1-26].

although few of these applications are documented in the literature. Sustained fast valving(sustained power reduction) may be necessary for a stable post-disturbance equilibrium.

AEP routinely reexamines the stability of the Rockport generation–transmission complexand the effectiveness of temporary fast valving. The Rockport Operating Guide is updatedto reflect changes in operating conditions, changes in controls or operating practices, andchanges in the regional transmission network. Figure 1-7 illustrates the effectiveness ofthe fast valving. The simulated operating conditions and event include a single prioroutage and a single phase fault, unsuccessfully cleared by single-phase switching at +50milliseconds, with successful backup three phase clearing 0.55 seconds after the fault.The plots are of the consequent changes in speed and rotor angle position. The upper plotsof Figure 1-7 are with temporary fast valving, and the lower plots are without fastvalving.

Generator tripping. Generator tripping is an effective and economic control especially ifhydro units are used. Tripping of fossil units, especially gas- or oil-fired units, may beattractive if tripping to house load is possible and reliable. Gas turbine and combined-cycle plants constitute a large percentage of new generation. Occasional tripping of theseunits is feasible and can become an attractive stability control in the future.

Most generator tripping controls are event-based (based on outage of generating plantout-going lines or outage of tie lines). Several advanced response-based generatortripping controls, however, have been implemented.

1-12

Fig. 1-7. Simulation of effect of temporary fast valving at Rockport for prior circuitoutage and single-phase fault with unsuccessful single-pole switching. Top plots are withfast valving and bottom plots are without fast valving.

The Acceleration Trend Relay (ATR) is implemented at the Colstrip generating plant ineastern Montana [1-28]. The plant consists of two 330 MW units and two 700 MW units.The microprocessor-based controller measures rotor speed and generator power, andcomputes acceleration and angle. Tripping of 16–100% of plant generation is based oneleven trip algorithms involving acceleration, speed and angle changes. Because of thelong distance to Pacific Northwest load centers, the ATR has operated many times, bothdesirably and undesirably. There are proposals to use voltage angle measurementinformation (Colstrip 500-kV voltage angle relative to Grand Coulee and other Northwestlocations) to adaptively adjust ATR settings, or as additional information for trip

1-13

algorithms. Another possibility is to provide speed or frequency measurements fromGrand Coulee and other locations to base algorithms on speed difference rather than onlythe local Colstrip speed [1-29].

A Tokyo Electric Power Company stabilizing control predicts generator angle changesand decides the minimum number of generators to trip [1-30]. Local generator electricpower, voltage and current measurements are used to estimate angles. The control hasworked correctly for several actual disturbances.

The Tokyo Electric Power Company is also developing an emergency control systemwhich uses a predictive prevention method for step-out of pumped storage generators [1-31,1-32]. In the new method, the generators in TEPCO’s network which swing againsttheir local pumped storage generators after serious fault are treated as an external powersystem. The parameters in the external system, such as angle and inertia, are estimated byusing local on-line information. The behavior of a local pumped storage generator ispredicted based on equations of motion. Control actions (the number of generators to betripped) are determined based on the prediction.

Reference 1-33 describes response-based generator tripping using a phase-planecontroller. The controller is based on the apparent resistance/rate of change of apparentresistance (R–Rdot) phase plane, which is closely related to an angle difference/speeddifference phase plane between two areas. The primary use of the controller is forcontrolled separation of the Pacific AC Intertie. Figure 1-8 shows simulation resultswhere 600 MW of generator tripping reduces the likelihood of controlled separation.

Fig. 1-8. R–Rdot phase plane for loss of Pacific HVDC Intertie (2000 MW). Solidtrajectory is without additional generator tripping. Dashed trajectory is with additional600 MW of generator tripping initiated by the R–Rdot controller generator trip switchingline [1-33].

1-14

Fast fault clearing, high-speed reclosing, and single-pole switching. Clearing time ofclose-in faults can be less than three cycles using conventional protective relays andcircuit breakers. Typical EHV circuit breakers have two-cycle opening time. One-cyclebreakers have been developed [1-34], but special breakers are seldom justified. Highmagnitude short circuits may be detected as fast as one-fourth cycle by non-directionalovercurrent relays. Ultra-high-speed traveling wave relays are also available [1-35]. Withsuch short clearing times, and considering that most EHV faults are single-phase, theremoved transmission lines or other elements may be the major contributor to generatoracceleration. This is especially true if non-faulted equipment is removed by “sympathetic”relaying.

High-speed three-pole reclosing is an effective method of improving stability andreliability. Reclosing is before the maximum of the first forward angular swing, but after30–40 cycle time for arc extinction. During a lightning storm, high speed reclosing keepsthe maximum number of lines in service. High-speed reclosing is effective whenunfaulted lines trip because of relay misoperations.

Unsuccessful high-speed reclosing into a permanent fault can cause instability, and canalso compound the torsional duty imposed on turbine-generator shafts. Solutions includereclosing only for single-phase faults, and reclosing from the weak end with hot-linechecking prior to reclosing at the generator end. Communication signals from the weakend indicating successful reclosing can also be used to enable reclosing at the generatorend [1-38].

Single-pole switching is a practical means to improve stability and reliability in EHVnetworks where most circuit breakers have independent pole operation [1-36,1-37].Several methods are used to ensure secondary arc extinction. For short lines, no specialmethods are needed. For long lines, the four-reactor scheme [1-39,1-40] is mostcommonly used. High-speed grounding switches may be used [1-41]. A hybrid reclosingmethod used by Bonneville Power Administration employs single-pole tripping, but withthree-pole tripping on the backswing followed by rapid three-pole reclosure; the three-pole tripping ensures secondary arc extinction [1-36].

Single-pole switching may necessitate positive sequence filtering in stability control inputsignals.

For advanced stability control, signal processing and pattern recognition techniques maybe developed to detect secondary arc extinction [1-42,1-43]. Reclosing into a fault isavoided and single-pole reclosing success is improved.

High-speed reclosing or single-pole switching may not allow increased power transfersbecause deterministic reliability criteria generally specifies permanent faults.Nevertheless, fast reclosing provides “defense-in-depth” for frequently occurring single-phase temporary faults and false operation of protective relays. The probability of powerfailures because of multiple line outages is greatly reduced.

Dynamic braking. Shunt dynamic brakes using mechanical switching have been usedinfrequently [1-2]. Normally the insertion time is fixed. One attractive method not

1-15

requiring switching is neutral-to-ground resistors in generator step-up transformers.Braking automatically results for ground faults — which are most common.

Often generator tripping, which helps ensure a post-disturbance equilibrium, is a bettersolution.

Thyristor switching of dynamic brakes has been proposed. Thyristor switching or phasecontrol minimizes generator torsional duty [1-44], and can be a subsynchronousresonance countermeasure [1-45].

Load tripping and modulation. Load tripping is similar in concept to generator trippingbut is at the receiving end to reduce deceleration of receiving-end generation.Interruptible industrial load is commonly used. For example, reference 1-46 describestripping of up to 3000 MW of industrial load following outages during power importconditions.

Rather than tripping large blocks of industrial load, it may be possible to trip low prioritycommercial and residential load such as space and water heaters, or air conditioners. Thisis less disruptive and the consumer may not even notice brief interruptions. The feasibilityof this control depends on implementation of direct load control as part of demand sidemanagement, and on the installation of high-speed communication links to consumerswith high-speed actuators at load devices. Although unlikely because of economics,appliances such as heaters could be designed to provide frequency sensitivity by localmeasurements.

Load tripping is also used for voltage stability. Here the communication and actuatorspeeds are generally not as critical.

It’s also possible to modulate loads such as heaters to damp oscillations [1-47–50]. Thisis described in Chapter 7.

Clearly load tripping or modulation of small loads will depend on the economics, and thedevelopment of fast communications and actuators.

Reactive power compensation switching or modulation. Controlled series or shuntcompensation improves stability, with series compensation generally being the most costeffective [1-86]. For switched compensation, either mechanical or power electronicswitches may be used. For continuous modulation, thyristor phase control of a reactor(TCR) is used. Mechanical switching has the advantage of lower cost. The operatingtimes of circuit breakers are usually adequate, especially for interarea oscillations.Mechanical switching is generally single insertion of compensation for synchronizingsupport. In addition to previously mentioned advantages, power electronic control hasadvantages in subsynchronous resonance performance [1-51].

For synchronizing support, high-speed series capacitor switching has been usedeffectively on the North American Pacific AC intertie for over 25 years [1-52]. The mainapplication is for full or partial outages of the parallel Pacific HVDC intertie (event-driven control using transfer trip over microwave radio). Series capacitors are inserted bycircuit breaker opening; operators bypass the series capacitors some minutes after the

1-16

event. Response-based control using an impedance relay was also used for some years,and new response-based controls are being investigated.

Thyristor-based series compensation switching or modulation has been developed withseveral installations in service or planned [1-10,1-53,1-54]. Thyristor-controlled seriescompensation (TCSC) allows significant time-current dependent increase in seriescapacitive reactance over the nominal reactance. With appropriate controls, this increasein reactance can be a powerful stabilizing force [1-55,1-56].

As described in Chapter 7, thyristor-controlled series compensation was chosen for the1020 km, 500-kV intertie between the Brazilian north/northeast networks and thesouth/southeast networks [1-57]. Also described in Chapter 7 is a TCSC application inChina for integration of a remote power plant using two parallel 500-kV transmissionlines (1300 km). Transient stability simulations indicate that 25% thyristor controlledcompensation is more effective than 45% fixed compensation. Several advanced TCSCcontrol techniques are promising [1-58].

For synchronizing support, high speed switching of shunt capacitor banks is alsoeffective. Again on the Pacific AC intertie, four 200 MVAr shunt banks are switched forHVDC and 500-kV ac line outages [1-16]; response-based controls based on voltage areinstalled.

High speed mechanical switching of shunt banks as part of a static var system is common.For example, the Forbes static var system near Duluth, Minnesota USA includes two 300MVAr 500-kV shunt capacitor banks [1-59]. Generally it’s cost-effective to augmentpower electronic controlled compensation with fixed or mechanically-switchedcompensation.

Static var compensators are applied along interconnections to improve synchronizing anddamping support. Voltage support at intermediate points allow operation at angles above90°. Reference 1-60 provides an example using five SVCs with only voltage control toimprove stability for a proposed interconnection of the Scandinavian (Nordel) and mainEuropean (UCPTE) power systems.

SVCs are modulated to improve oscillation damping. One study [1-1,1-61] showed linecurrent magnitude to be the most effective input signal. Synchronous condensers canprovide similar benefits, but nowadays are usually not competitive with power electronicequipment. Available SVCs in load areas may be used to indirectly modulate load toprovide synchronizing or damping forces.

Digital controls allow many new control strategies. Gain supervision and optimizationadaptive control is common. For series or shunt power electronic devices, control modeselection allows bang-bang control, synchronizing versus damping control, and other non-linear and adaptive strategies.

Current injection by voltage source converters. Advanced power electronic controlledequipment employ gate turn-off thyristors or other devices with current turnoff capability.Reactive power injection devices include static compensator (STATCOM), static

1-17

synchronous series compensator (SSSC), and unified power flow controller (UPFC).Reference 1-1 describes use of these devices for oscillation damping.

As with conventional thyristor-based equipment, it’s often effective for voltage sourceinverter control to also coordinate mechanical switching.

Voltage source inverters may also be used for real power series or shunt injection.Superconducting magnetic energy storage (SMES) or battery storage is the most common.For angle stability control, injection of real power is more effective than reactive power.SMES or battery storage provides both active and reactive power control.

For transient stability improvement, SMES can be of smaller MVA size and possiblylower cost than a STATCOM. SMES may be less location dependent than a STATCOM.

Fast voltage phase angle control. Voltage phase angles and thereby rotor angles can berapidly controlled by power electronic controlled series compensation (discussed above)or phase shifting transformers. This provides powerful stability control. Although onetype of thyristor-controlled phase shifting transformer was developed almost twenty yearsago [1-62], high cost has presumably prevented installations. Reference 1-63 describes anapplication study.

The unified power flow controller incorporates GTO-thyristor phase shifting and seriescompensation control, and one installation (not a transient stability application) is inservice [1-53].

One concept employs power electronic series or phase shifting equipment to directlycontrol angles across an interconnection within a small range [1-64]. On a power–anglecurve, this can be visualized as keeping high synchronizing coefficient (slope of power–angle curve) during disturbances.

Bonneville Power Administration developed a novel method for transient stability byhigh speed 120° phase rotation of transmission lines between networks losingsynchronism [1-54]. This technique is very powerful (perhaps too powerful!) and raisesreliability and robustness issues especially in the usual case where several lines form theinterconnection. It has not been implemented.

HVDC link supplementary controls. HVDC dc links are installed for power transferreasons. In contrast to the above power electronic devices, the available HVDCconverters provide the actuators so that stability control is inexpensive. For long distanceHVDC links within a synchronous network, HVDC modulation can provide powerfulstabilization, with active and reactive power injections at each converter. Controlrobustness, however, is a concern [1-1,1-7].

References 1-1, 1-65–67 and 1-87 describe HVDC link stability controls. The PacificHVDC Intertie modulation control, implemented in 1976, is unique in that a remote inputsignal from the parallel Pacific AC Intertie was used. Figure 1-9 shows commissioningtest results.

Adjustable-speed (doubly-fed) generation. References 1-1, 1-68, 1-69, and Appendix Adescribe stability benefits of adjustable speed synchronous machines that have been

1-18

Fig. 1-9. System response to Pacific AC Intertie series capacitor bypass with and withoutdc modulation [1-66].

developed for pumped storage applications. Control of excitation frequency enables directcontrol of rotor angle. Since the frequency converter only supplies power to the rotor, thecost may be low enough to be competitive with alternatives. Reference 1-88 describesdoubly-fed turbo-generators.

Controlled separation and underfrequency load shedding. For very severedisturbances and failures, maintaining synchronism may not be possible or cost-effective.Controlled separation (islanding) based on out-of-step detection or parallel path outagesmitigates the effects of instability. The generation/load imbalances in the islands that areformed should be small enough that the islands stabilize. Undesirable generation trippingduring voltage and frequency swings must be minimized through adequate control andprotection design and settings. Underfrequency load shedding may be required in islandsthat were importing power.

References 1-33, 1-70, and 1-71 describe advanced controlled separation schemes. Recentproposals advocate use of voltage phase angle measurements for controlled separation.

1.4 Dynamic Security AssessmentControl design and settings, along with transfer limits, are usually based on off-linesimulation (time and frequency domain), and on field tests. Controls must then operateappropriately for a variety of operating conditions and disturbances.

Recently, however, on-line dynamic (or transient) stability/security assessment softwarehas been developed. State estimation and on-line power flow monitoring provide the baseoperating conditions. Simulation of potential disturbances is then based on actualoperating conditions, reducing uncertainty of the control environment. Dynamic securityassessment is presently used to determine arming levels for generator tripping controls [1-72,1-73].

With today’s computer capabilities, hundreds or thousands of large-scale simulations maybe run each day to provide an organized database of system stability properties. Securityassessment is made efficient by techniques such as fast screening and contingency

1-19

selection, and smart termination of strongly stable or unstable cases. Parallel computationis straightforward using multiple workstations for different simulation cases; commoninitiation may be used for the different contingencies

In the future, dynamic security assessment may be used for control adaptation to currentoperating conditions. Another possibility is stability control based on neural network ordecision tree pattern recognition. Dynamic security assessment provides the database forpattern recognition techniques. Pattern recognition may be considered data compressionof security assessment results.

Industry restructuring requiring near real-time power transfer capability determinationmay accelerate the implementation of dynamic security assessment, facilitating advancedstability controls.

We further describe on-line security assessment in Chapter 5.

1.5 Intelligent ControlsMention has already been made of rule-based controls and pattern recognition basedcontrols. Fuzzy logic may be used for rule-based control.

As a possibility, reference 1-74 describes a sophisticated self-organizing neural fuzzycontroller (SONFC) based on the speed–acceleration phase plane. Compared to theangle–speed phase plane, control tends to be faster and both final states are zero (usingangle, the post-disturbance equilibrium angle is not known in advance). The controllersare located at generator plants. Acceleration and speed can be easily measured/computedusing, for example, the techniques developed for power system stabilizers.

The SONFC could be expanded to incorporate remote measurements. Dynamic securityassessment simulations could be used for updating/retraining of the neural network fuzzycontroller. The SONFC is suitable for generator tripping, series or shunt capacitorswitching, HVDC control, etc.

We further describe intelligent controls in Chapter 4.

1.6 Effect of Industry Restructuring on Stability ControlsIndustry restructuring will have many impacts on power system stability. New, frequentlychanging power transfer patterns cause new stability problems. Most stability and transfercapability problems must be solved by new controls and new substation equipment, ratherthan by new transmission lines [1-75].

Different ownership of generation, transmission and distribution makes necessary powersystem engineering more difficult. New power industry standards along with ancillaryservices mechanisms are being developed. Controls such as generator or load tripping,fast valving, higher than standard exciter ceilings, and power system stabilizers may beancillary services. In large interconnections, independent grid operators or securitycoordination centers may facilitate dynamic security assessment and centralized stabilitycontrols.

1-20

We further describe the effect of industry restructuring on stability controls in Chapter 8.

1.7 Experience from Recent Power FailuresRecent cascading power outages demonstrated the impact of control and protectionfailures, the need for “defense-in-depth” or “multiple lines of defense,” and the need foradvanced stability controls.

The July 2, 1996 and August 10, 1996 power failures [1-76–80] in western NorthAmerica showed need for improvements and innovations in stability control areas suchas:

• Fast insertion of reactive power compensation for voltage support, and fast generatortripping using response-based controls.

• HVDC, TCSC, and SVC control for stability.

• Power system stabilizer design and tuning.

• Controlled separation.

• Power system modeling and data validation for control design.

• Control adaptation to actual operating conditions.

Figure 1-10 shows the development of the August 10 breakup

Other blackouts have occurred recently in the North American Upper Midwest [1-80],and in Brazil. In Brazil, new emergency controls for generator/load tripping andcontrolled separation are being added.

Defense-in-depth/multiple line of defense for system reliability includes risk managementin system operation (e.g., reduced power transfers during storm conditions), fast andreliable protective relaying, high-speed three or single pole reclosing, best practice localstability controls (e.g., thyristor exciters with PSS). The final lines of defense mitigate theeffects of extreme disturbances, and may include generator/load tripping, controlledseparation, and underfrequency or undervoltage load shedding.

1.8 Coordination with other CIGRÉ and Industry WorkReferences 1-1, 1-81, and 1-82 document recent CIGRÉ and IEEE work related to anglestability control. These works are valuable, providing comprehensive description of manyaspects of stability. CIGRÉ TF 38.02.16, Impact of Interactions among Power SystemControls, CIGRÉ TF 38.02.19, System Protection in the Power System: modeling andanalysis, and CIGRÉ TF 38.02.20 Advanced Power System Controls Using IntelligentSystems, are currently underway.

Our intent is to complement rather than duplicate other industry work.

1-21

[bitmap version

200 300 400 500 600 700 8001100

1200

1300

1400

1500

008 Malin-Round Mountain #1 MWcaseID=Aug10E5loadPF casetime =04/16/98_14:41:48

Time in Seconds

0.264 Hz,3.46% damping

0.252 Hz0.276 Hz

15:42:03Keeler-Allston line trips

15:48:51Out-of-Step separation

15:47:36Ross-Lexington line trips/McNary generation drops off

Reference time = 15:35:30 PDT

Fig. 1-10. Power flow on Oregon–California 500-kV line during initial portion of August10, 1996 breakup. Following separation of the Pacific AC intertie, uncontrolledseparations broke the system into four islands with loss of 30,489 MW of load.

1.9 SummaryPower system angle stability can be improved by a wide variety of controls. Somemethods have been used effectively for many years, both at generating plants and intransmission networks. New control techniques and actuating equipment are promising.

This chapter provides a broad survey of available stability control techniques withemphasis on new and emerging technology. The following chapters provide in-depthevaluation of the many issues in the selection and design of stability controls.

References1-1 CIGRÉ TF 38.01.07, Analysis and Control of Power System Oscillations,

Brochure 111, December 1996.

1-2 P. Kundur, Power System Stability and Control, McGraw-Hill, 1994.

1-3 IEEE Special Stability Controls Working Group, “Annotated Bibliography onPower System Stability Controls: 1986-1994,” IEEE Transactions on PowerSystems, Vol. 11, No. 2, pp. 794-800, August 1996.

1-22

1-4 J. F. Christensen, “New Control Strategies for Utilizing Power System NetworkMore Effectively,” Electra, No. 173, pp. 5–16, August 1997.

1-5 IEEE Discrete Supplementary Control Task Force, “A Description of DiscreteSupplementary Controls for Stability,” IEEE Transactions on Power Apparatusand Systems, Vol. PAS-97, pp. 149–165, January/February 1978.

1-6 A. Hammad, “Stability and Control of HVDC and AC Transmission in Parallel,”IEEE Transactions on Power Delivery, Vol. 14, No. 4, October 1999.

1-7 J. F. Hauer, “Robust Damping Controls for Large Power Systems,” IEEE ControlSystems Magazine, January 1989.

1-8 J. F. Hauer, “Robustness Issues in Stability Control of Large Electric PowerSystems,” 32nd IEEE Conference on Decision and Control, San Antonio, Texas,December 15–17, 1993.

1-9 T. Studt, “Computer Scientists Search for Ties to Biological Intelligence,” R&DMagazine, pp. 77–78, October 1998.

1-10 North American Electric Reliability Council, NERC Planning Standards,September 1997 (available at www.nerc.com).

1-11 A. F. Djakov, A. Bondarenko, M. G. Portnoi, V. A. Semenov, I. Z. Gluskin, V. D.Kovalev, V. I. Berdnikov, and V. A. Stroev, “The Operation of Integrated PowerSystems Close to Operating Limits with the Help of Emergency Control Systems,”CIGRÉ, paper 39-109, 1998.

1-12 IEEE/ CIGRÉ Committee Report (P. M. Anderson and B. K. LeReverend),“Industry Experience with Special Protection Schemes,” IEEE Transactions onPower Systems, Vol. 11, No. 3, pp. 1166–1179, August 1996.

1-13 D. Dodge, W. Doel, and S. Smith, “Power System Stability Control Using FaultTolerant Technology,” ISA Instrumentation in Power Industry, Vol. 33, 33rdPower Instrumentation Symposium, May 21–23, 1990, paper 90-1323.

1-14 CIGRÉ Task Force 13.00.1, “Controlled Switching — A State-of-the-Art Survey,”Electra, No. 163, pp. 65–97, December 1995.

1-15 J. H. Brunke, J. H. Esztergalyos, A. H. Khan, and D. S. Johnson, “Benefits ofMicroprocessor-Based Circuit Breaker Control,” CIGRÉ, paper 23/13-10, 1994.

1-16 B. C. Furumasu and R. M. Hasibar, “Design and Installation of 500-kV Back-to-Back Shunt Capacitor Banks,” IEEE Transactions on Power Delivery, Vol. 7, No.2, pp. 539–545, April 1992.

1-17 K. B. Cho, J. B. Kim, E. B. Shim, and J. W. Park, “Development of an IntelligentAutoreclosing Concept using Neuro-Fuzzy Techniques — An Optimal ControlledSwitching for System Operation,” CIGRÉ, paper 13-114, 1998.

1-18 T. Torizuka and H. Tanaka, “An Outline of Power System Technologies inJapan,” Electric Power Systems Research, No. 44, pp. 1–5, 1998.

1-23

1-19 IEEE Digital Excitation Applications Task Force, “Digital Excitation Technology— A Review of Features, Functions and Benefits,” IEEE Transactions on EnergyConversion, Vol. 12, No. 3, September 1997.

1-20 K. E. Bollinger, L. Nettleton, T. Greenwood-Madsen, and M. Salyzyn,“Experience with Digital Power System Stabilizers at Steam and HydroGenerating Plants,” IEEE Transactions on Energy Conversion, Vol. 8, No. 2, June1993.

1-21 L. M. Hajagos, and G. R. Gerube, “Utility Experience with Digital ExcitationSystems,” IEEE Transactions on Power Systems, Vol. 13, No. 1, pp. 165–170,February 1998.

1-22 V. Arcidiancone, S. Corsi, G. Ottaviani, S. Togno, G. Baroffio, C. Raffaelli, andE. Rosa, “The ENEL’s Experience on the Evolution of Excitation Control Systemsthrough Microprocessor Technology,” IEEE Transactions on Energy Conversion,Vol. 13, No. 3, pp. 292–299, September 1998.

1-23 A. S. Rubenstein and W. W. Walkley, “Control of Reactive KVA with ModernAmplidyne Voltage Regulators,” AIEE Transactions, pp. 961–970, December1957.

1-24 A. S. Dehdashti, J. F. Luini, and Z. Peng, “Dynamic Voltage Control by RemoteVoltage Regulation for Pumped Storage Plants,” IEEE Transactions on PowerSystems, Vol. 3, No. 3, pp. 1188–1192, August 1988.

1-25 D. C. Lee and P. Kundur, “Advanced Excitation Controls for Power SystemStability Enhancement,” CIGRÉ, paper 38-01, 1986.

1-26 C. W. Taylor, J. R. Mechenbier, and C. E. Matthews, “Transient ExcitationBoosting at Grand Coulee Third Power Plant,” IEEE Transactions on PowerSystems, Vol. 8, No. 3, pp. 1291–1298, August 1993.

1-27 N. B. Bhatt, “Field Experience with Momentary Fast Turbine Valving and OtherSpecial Stability Controls Employed at AEP’s Rockport Plant,” IEEETransactions on Power Systems, Vol. 11, No. 1, pp. 155–161, February 1996.

1-28 C. A. Stigers, C. S. Woods, J. R. Smith, and R. D. Setterstrom, “The AccelerationTrend Relay for Generator Stabilization at Colstrip,” IEEE Transactions on PowerDelivery, Vol. 12, No. 3, pp. 1074–1081, July 1997.

1-29 D. N. Kosterev, J. Esztergalyos, and C. A. Stigers, “Feasibility Study of UsingSynchronized Phasor Measurements for Generator Dropping Controls in theColstrip System,” IEEE Transactions on Power Systems, Vol. 13, No. 3, pp. 755–762, August 1998.

1-30 K. Matsuzawa, K. Yanagihashi, J. Tsukita, M. Sato, T. Nakamura, and A.Takeuchi, “Stabilizing Control System Preventing Loss of Synchronism fromExtension and Its Actual Operating Experience,” IEEE Transactions on PowerSystems, Vol. 10, No. 3, pp. 1606–1613, August 1995.

1-24

1-31 Y. Kojima, H. Taoka, H. Oshida, and T. Goda, “On-line Modeling for EmergencyControl Systems,” IFAC/CIGRE Symposium on Control of Power Systems andPower Plant, pp. 627–632, 1997.

1-32 S. Imai, T. Syoji, K. Yanagihashi, Y. Kojima, Y. Kowada, H. Oshida, and T.Goda, “Development of Predictive Prevention Method for Mid-term StabilityProblem using Only Local Information,” Transactions of IEE Japan, Vol. 118-B,No.9, 1998.

1-33 J. M. Haner, T. D. Laughlin, and C. W. Taylor, “Experience with the R-Rdot Out-of-Step Relay,” IEEE Transactions on Power Delivery, Vol. PWRD-1, No. 2, pp.35–39, April 1986.

1-34 R. O. Berglund, W. A. Mittelstadt, M. L. Shelton, P. Barkan, C. G. Dewey, and K.M. Skreiner, “One-Cycle Fault Interruption at 500 kV: System Benefits andBreaker Designs,” IEEE Transactions on Power Apparatus and Systems, Vol.PAS-93, pp. 1240–1251, September/October 1974.

1-35 J. H. Esztergalyos, M. T. Yee, M. Chamia, and S. Lieberman, “The Developmentand Operation of an Ultra High Speed Relaying System for EHV TransmissionLines,” CIGRÉ, paper 34-04, 1978.

1-36 IEEE Committee Report, “Single-Pole Switching for Stability and Reliability,”IEEE Transactions on Power Systems, Vol. PWRS-1, pp. 25–36, May 1986.

1-37 A. K. Belotelov, A. F. Dyakov, G. G. Fokin, V. V. Ilynichnin, A. I. Leviush, andV. M. Strelkov, “Application of Automatic Reclosing in High Voltage Networksof the UPG of Russia Under New Conditions,” CIGRÉ, paper 34-203, 1998.

1-38 K. C. Behrendt, “Relay-to-Relay Digital Logic Communication for LineProtection, Monitoring, and Control,” Proceedings of the 23rd Annual WesternProtective Relay Conference, Spokane, Washington USA, October 1996.

1-39 N. Knutsen, “Single-Phase Switching of Transmission lines Using Reactors forExtinction of the Secondary Arc,” CIGRÉ, paper 310, 1962.

1-40 E. W. Kimbark, “Suppression of Ground-Fault Arcs on Single-Pole SwitchedLines by Shunt Reactors,” IEEE Transactions on Power Apparatus and Systems,Vol. PAS-83, No. 3, pp. 285–290, March 1964.

1-41 R. M. Hasibar, A. C. Legate, J. H. Brunke, and W. G. Peterson, “The Applicationof High-Speed Grounding Switches for Single-Pole Reclosing on 500-kV PowerSystems, IEEE Transactions on Power Apparatus and Systems, Vol. PAS-100,No. 4, pp. 1512–1515, April 1981.

1-42 D. S. Fitton, R. W. Dunn, R. K. Aggarwal, A. T. Johns, and A. Bennett, “Designand Implementation of an Adaptive Single Pole Autoreclosure Technique forTransmission Lines using Artificial Neural Networks,” IEEE Transactions onPower Delivery, Vol. 11, No. 2, pp. 748–756, April 1996.

1-25

1-43 M. B. Djuric and V. V. Terzija, “A New Approach to the Arcing Faults Detectionfor Fast Autoreclosure in Transmission Systems,” IEEE Transactions on PowerDelivery, Vol. 10, No. 4, pp. 1793–1798, October 1995.

1-44 W. Bayer, K. Habur, D. Povh, D. A. Jacobson, J. M. G. Guedes, and D. A.Marshall, “Long Distance Transmission with Parallel AC/DC Link From CahoraBassa (Mozambique) to South Africa and Zimbabwe,” CIGRÉ, paper 14-306,1996.

1-45 M. K. Donnelly, J. R. Smith, R. M. Johnson, J. F. Hauer, R. W. Brush, and R.Adapa, “Control of a Dynamic Brake to Reduce Turbine-Generator ShaftTransient Torques,” IEEE Transactions on Power Systems, Vol. 8 No. 1, pp. 67–73, February 1993.

1-46 C. W. Taylor, F. R. Nassief, and R. L. Cresap, “Northwest Power Pool TransientStability and Load Shedding Controls for Generation–Load Imbalances,” IEEETransactions on Power Apparatus and Systems, Vol. PAS-100, No. 7, pp. 3486–3495, July 1981.

1-47 O. Samuelsson and B. Eliasson, “Damping of Electro-Mechanical Oscillations in aMultimachine System by Direct Load Control,” IEEE Transactions on PowerSystems, Vol. 12, No. 4, pp. 1604–1609, November 1997.

1-48 I. Kamwa, R. Grondin, D. Asber, J. P. Gingras, and G. Trudel, “Active PowerStabilizers for Multimachine Power Systems: Challenges and Prospects,” IEEETransactions on Power Systems, Vol. 13, No. 4, pp. 1352–1358, November 1998.

1-49 J. Dagle, “Distributed-FACTS: End-Use Load Control for Power System DynamicStability Enhancement,” EPRI Conference, The Future of Power Delivery in the21st Century, 18–20 November 1997, La Jolla, California.

1-50 I. Kamwa, R. Grondin, D. Asber, J. P. Gingras, and G. Trudel, “Large-ScaleActive-Load Modulation for Angle Stability Improvement,” IEEE Transactionson Power Systems, Vol. 14, No. 2, pp. 582–590, May 1999.

1-51 A. Hammad and M. El-Sadek, “Application of a Thyristor Controlled VArCompensator for Damping Sub-synchronous Oscillations in Power Systems,”IEEE Trans., Vol. PAS-103, No. 1, pp 198–212, January 1984.

1-52 E. W. Kimbark, “Improvement of System Stability by Switched SeriesCapacitors,” IEEE Transactions on Power Apparatus and Systems, Vol. PAS-85,No. 2, pp. 180–188, February 1966.

1-53 M. Rahman, M. Ahmed, R. Gutman, R. J. O’Keefe, R. J. Nelson, and J. Bian,“UPFC Application on the AEP System: Planning Considerations,” IEEETransactions on Power Systems, Vol. 12, No. 4, pp. 1695–1701, November 1997.

1-54 R. L. Cresap, C. W. Taylor, and M. J. Kreipe, “Transient Stability Enhancementby 120-Degree Phase Rotation,” IEEE Transactions on Power Apparatus andSystems, Vol. PAS-100, pp. 745–753, February 1981.

1-26

1-55 N. Christl, K. Sadek, R. Hedin, P. Lützelberger, P. E. Krause, A. H. Montoya, S.M. McKenna, and D. Torgerson, “Advanced Series Compensation (ASC) withThyristor Controlled Impedance, CIGRÉ, paper 14/37/38-05, 1992.

1-56 R. J. Piwko, C. A. Wegner, B. C. Furumasu, B. L. Damsky, and J. D. Eden, “TheSlatt Thyristor-Controlled Series Capacitor Project — Design, Installation,Commissioning and System Testing,” CIGRÉ, paper 14-104, 1994.

1-57 C. Gama, R. L. Leoni, J. Gribel, R. Fraga, M.J. Eiras, W. Ping, A. Ricardo, J.Cavalcanti, and R. Tenório, “Brazilian North–South Interconnection —Application of Thyristor Controlled Series Compensation (TCSC) to Damp Inter-Area Oscillation Mode,” CIGRÉ, paper 14-101, 1998.

1-58 X. Zhou et al., “Analysis and Control of Yimin–Fentun 500 kV TCSC System,”Electric Power Systems Research 46 (1998), pp. 157–168.

1-59 G. Sybille, P. Giroux, S. Dellwo, R. Mazur, and G. Sweezy, “Simulator and FieldTesting of Forbes SVC,” IEEE Transactions on Power Delivery, Vol. 11, No. 3,pp. 1507–1514, July 1996.

1-60 J. F. Christensen and T. Østrup, “ Example of Reinforcing an AC InterconnectionBetween Two Large Networks by Use of SVC,” Electra, No. 132, October 1990.

1-61 E. V. Larsen and J. H. Chow, “SVC Control Design Concepts for SystemDynamic Performance,” Application of Static Var Systems for System DynamicPerformance, IEEE Special Publication 87TH1087-5-PWR, pp. 36–53, 1987.

1-62 H. Stemmler and G. Güth, “The Thyristor-Controlled Static Phase Shifter—ANew Tool for Power Flow Control in AC Transmission Systems,” Brown BoveriReview, Vol. 69, No. 3, pp. 73–78, March 1982.

1-63 Y. J. Fang and D. C. Macdonald, “Dynamic Quadrature Booster as an Aid toSystem Stability, IEE Proc.—Gener. Transm. Distrib., Vol. 145, No. 1, pp. 4147,January 1998.

1-64 J. Hauer, D. Trudnowski, G. Rogers, W. Mittelstadt, W. Litzenberger, and J.Johnson, “Keeping an Eye on Power System Dynamics,” IEEE ComputerApplications in Power, Vol. 10, No. 1, pp. 26–30, January 1997.

1-65 IEEE Committee Report, “HVDC Controls for System Dynamic Performance,”IEEE Transactions on Power Systems, Vol. 6, No. 2, pp. 743–752, May 1991.

1-66 R. L. Cresap, D. N. Scott, W. A. Mittelstadt, and C. W. Taylor, “OperatingExperience with Modulation of the Pacific HVDC Intertie,” IEEE Transactions onPower Apparatus and Systems, Vol. PAS-98, pp. 1053–1059, July/August 1978.

1-67 R. L. Cresap, D. N. Scott, W. A. Mittelstadt, and C. W. Taylor, “Damping ofPacific AC Intertie Oscillations via Modulation of the Parallel Pacific HVDCIntertie,” CIGRE 14-05, 1978.

1-27

1-68 W. Gish, J. Schurz, B. Milano, and F. Schleif, “An Adjustable SpeedSynchronous Machine For Hydroelectric Power Applications,” IEEE Transactionson Power Apparatus and Systems, Vol. 100, No. 5, pp.2171–2176, May 1981.

1-69 Y. Shakarian, L. Mamilo-Niants, Y. Vinitsky, A. Lochmatov, J. M. Kauffmann,H. Mangel, J. Deuse, K. Karoui, and J. Dubois, “Role of Two-Axis ExcitationGenerators in Power Systems,” CIGRÉ, paper 11/37-04, 1998. See also discussionof Joint Session 11/37 in Electra, No. 182, February 1999.

1-70 Y. Ohura, M. Suzuki, K. Yanagihashi, M. Yamaura, K. Omata, T. Nakamura, S.Mitamura, and H. Watanabe, “A Predictive Out-of-Step Protection System Basedon Observation of the Phase Difference Between Substations,” IEEE Transactionson Power Delivery, Vol. 5, No. 4, pp. 1695–1704, November 1990.

1-71 V. Centeno, A. G. Phadke, A. Edris, J. Benton, M. Gaudi, and G. Michel, “AnAdaptive Out of Step Relay,” IEEE Transactions on Power Delivery, Vol. 12, No.1, pp. 61–71, January 1997.

1-72 Y. Mansour, E. Vaahedi, A. Y. Chang, B. R. Corns, B. W. Garrett, K. Demaree, T.Athay, and K. Cheung, “B. C. Hydro’s On-line Transient Stability Assessment(TSA) Model Development, Analysis, and Post-processing,” IEEE Transactionson Power Systems, Vol. 10, No. 1, pp. 241–253, February 1995.

1-73 H. Ota, Y. Kitayama, H. Ito, N. Fukushima, K. Omata, K. Morita, and Y. Kokai,“Development of Transient Stability Control System (TSC System) Based on On-Line Stability Calculation,” IEEE Transactions on Power Systems, Vol. 11 No. 3,pp. 1463–1472, August 1996.

1-74 H. C. Chang and M. H. Wang, “Neural Network-Based Self-Organizing FuzzyController for Transient Stability of Multimachine Power Systems,” IEEETransactions on Energy Conversion, Vol. 10, No. 2, pp. 339–347, June 1995.

1-75 K. N. Mortensen, “Transmission System Open Access Versus System Reliability:A Case History—the MAPP Disturbance of June 25, 1998,” Proceedings ofIEEE/PES 1999 Winter Meeting, pp. 1281–1284.

1-76 WSCC reports on July 2, 1996 and August 10, 1996 outages — available atwww.wscc.com.

1-77 C. W. Taylor and D. C. Erickson, “Recording and Analyzing the July 2 CascadingOutage,” IEEE Computer Applications in Power, Vol. 10, No. 1, pp. 26–30,January 1997.

1-78 D. N. Kosterev, C. W. Taylor, and W. A. Mittelstadt, “Model Validation for theAugust 10, 1996 WSCC System Outage,” IEEE Transactions on Power Systems,Vol. 14, No. 3, pp. 967–979, August 1999.

1-79 R. Bunch and D. N. Kosterev, “Design and Implementation of AC VoltageDependent Current Order Limiter at Pacific HVDC Intertie,” to appear in IEEETransactions on Power Systems.

1-28

1-80 C. W. Taylor, “Improving Grid Behavior,” IEEE Spectrum, pp. 40–45, June 1999.

1-81 CIGRÉ TF 38.01.06 Load Flow Control in High Voltage Systems Using FACTSControllers, October 1995.

1-82 IEEE FACTS Working Group, FACTS Applications, IEEE/PES 96TP116-0.

1-83 T. Kuwabara, A. Shibuya, H. Furuta, E. Kita, and K. Mitsuhashi, “Design andDynamic Response Characteristics of 400 MW Adjustable Speed Pumped StorageUnit for Ohkawachi Power Station,” IEEE Transactions on Energy Conversion,Vol. 11, No. 2, pp. 376–384, June 1996.

1-84 K. Kudo, “Japanese Experience With a Converter-fed Variable Speed PumpedStorage System,” The International Journal on Hydropower & Dams, Vol. 1, No.2, March 1994.

1-85 D. Schäfer, B. Willy, and J. J. Simond, “Adjustable Speed Asynchronous Machinein Hydro Power Plants and its Advantages for Electric Grid Stability,” CIGRÉ,paper 11/37-01, 1998. See also discussion of Joint Session 11/37 in Electra, No.182, February 1999.

1-86 M. Noroozian and G. Andersson, “Damping of Power System Oscillations by Useof Controllable Components,” IEEE Transactions on Power Delivery, Vol. 9, No.4, pp. 2046–2054, October 1994.

1-87 T. Smed and G. Andersson, “Utilising HVDC to Damp Power Oscillations,” IEEETransactions on Power Delivery, Vol. 8, No. 2, pp. 620–627, April 1993.

1-88 Y. Shakaryan, Z. Hvoschinskaya, A. Lochmatov, L. Mamikonyants, Y. Vinitsky,J. M. Kauffmann, H. Mange, J. Deuse, J. Dubois, and K. Karoui, “Role of Two-Axis Excitation Turbo-Generators in Power Systems,” CIGRÉ, paper 11/37-04,1998.

Chapter 2

Advanced Linear and Nonlinear Control Design

Rotor angle stability of a power system, as introduced in Chapter 1, concerns theelectromechanical dynamics of generator rotors [2-1,2-2]. The rotors of all connected ACgenerators must operate at the same synchronous speed. Small oscillations betweengenerator rotors occur frequently. Changes in the rotor angle relationship betweengenerators are a function of generator loading, the distribution of loads in the network andthe topology of the electrical network. The rotor angles between generators normallychange very slowly as the system changes operating point through daily, weekly andseasonal cycles. Short-term transients will occur following a disturbance to the powersystem, and oscillations may arise from slowly evolving operating conditions.

As is the case with essentially all physical phenomena, power system rotor angle stabilityis inherently a nonlinear control problem. The control problem in power system anglestability has several additional complicating factors. These include:

• An accurate mathematical representation of an interconnected power system is usuallyof very high order, often containing several thousand state variables;

• The system is multivariable, often containing numerous generators each with theirown controllers;

• The system is continuously time varying, with daily and seasonal cycles as wellsudden short term changes;

• The system often contains significant levels of noise due partly to the constantchanging of many loads;

• The system contains numerous nonlinearities, including saturation of generators,exciters, nonlinear power transfer characteristics and nonlinear load characteristics;

• An interconnected power system covers a large geographic area, which may makecommunication and monitoring of the system difficult and expensive.

Despite all of these difficulties, many aspects of the control problem are addressable interms of the vast amount of theoretical information applicable to linear, or linearizable,systems.

When a disturbance impacts a generator’s mechanical and electrical torque balance, therotor of that machine must either speed up or slow down. The electrical torque will oftenchange more rapidly than the mechanical torque input because it is dependent upon theelectrical network variables which can change rapidly. These variables include the powertransmission capacity of the network and the state of all other machine rotors in thesystem. The changes in electrical torque within a generator can be resolved into twocomponents, one in phase with rotor angle and the other in phase with rotor speed. Thesecomponents are often referred to respectively as synchronizing and damping torques [2-3]. These concepts can be generalized in terms of state space modeling [2-68]. Theseconcepts illustrate two separate aspects of the rotor angle stability problem. A lack of

2-2

synchronizing torque often leads to rotor angle instability in the first swing of thegenerator rotor. Synchronizing torque is restored by fast acting control actions. Theproblem is referred to as the transient stability problem. The control actions include faultclearing, network reconfigurations, generator fast valving [2-4], resistive breaking, orgenerator tripping [2-5–7] and they often do not utilize feedback. Some recent workshowever have incorporated feedback control for fast valving [2-8].

In contrast, control designs to enhance damping torque usually rely on applications oftheory from linear feedback control design, and occasionally nonlinear feedback controldesign. These control designs deal primarily with small disturbance stability described interms of linear control concepts such as eigenvalues, poles and zeros, bode plots, anddamping. Some common control actuators for small signal stability are generatorexcitation systems including power system stabilizers (PSS), power electronic devices,modulated loads, and HVDC links. Reference 2-53 provides an overview of rotor anglestability related to PSS. Some recent works have discussed the plausibility of usingfeedback control to modulate the system loads to improve damping [2-9,2-10]. In anydisturbance both the synchronizing torque and damping torque aspects of the rotor anglestability problem exist. Small-disturbance stability (damping torques) must always exist,and large-disturbance stability (synchronizing torques) should exist for most severedisturbances.

Knowledge of the stability conditions of an interconnected power system is vital forreliable operation. The availability and proper design of stability controls cansignificantly extend the safe operating limits of interconnected power systems.

2.1 Nonlinear ControlAlthough power systems are inherently nonlinear most of the control design used inpractice is based on linear control theory. In recent years, however, there have beenseveral advances in the application of nonlinear control theory. The main impetus is toobtain more effective controllers by having the design account for the systemnonlinearities. References 2-11 and 2-67 discuss aspects of the nonlinear nature of powersystems.

Feedback linearization. One approach involves feedback linearization. The nonlineardynamics of the system are transformed into a linear (or partially linear) system so thatlinear control techniques can be used. References 2-12 and 2-42 discuss theoreticalaspects. The result is a transformation or an input signal that contains a nonlinear as wellas a linear component. This approach has been applied to power systems to controlgenerator power [2-8,2-13]. Both papers illustrate a significant improvement in dampingand transient stability of the power system when the mechanical power input to thegenerator can be effectively controlled. Reference 2-54 presents the application offeedback linearization to excitation control for angle stability of a multi-machine system.Reference 2-66 describes feedback linearization applied to a small parallel AC/DC testsystem for the enhancement of transient stability.

Adaptive control. Some or all of the nonlinearities are treated in terms of time varyingchanges in the system. As the system changes its operating point, a model of the system

2-3

can be determined and the control applied according to information about the model orthe deviation of the system from the model. There are many approaches to adaptivecontrol. Some conventional adaptive controllers have been applied to power systemproblems [2-14–16]. There are also adaptive control approaches involving fuzzy systemsand or neural nets [2-17–19].

Cost function. This approach to nonlinear control design involves the use of a costfunction or penalty function to evaluate the effectiveness of controller parameters for agiven control structure. In reference 2-20, a simple quadratic cost function is used toevaluate controller design parameters for a TCSC. The method involves a large numberof simulation studies to determine the best set of design parameters for a set of operatingconditions and expected disturbances. This approach is only feasible when the number ofdesign parameters to be determined is relatively small.

Discontinuous control. Discontinuous controls or “bang-bang” controls are the mostcommonly used emergency measures for maintaining transient stability when largedisturbances occur in a power network. Examples include generator tripping, seriescapacitor switching, generator excitation boosting, and dynamic braking. Theseapproaches are very effective in mitigating disturbances and maintaining rotor anglestability during the first swing of the rotor angles. The basic problem in most of thesestrategies is to determine the appropriate level of control action and the correct timing forthe switching actions. In general this is a nonlinear control problem. In some cases theproblem may reduce to being able to detect the appropriate conditions and begin thecontrol sequence. Many of these approaches rely on detailed and extensive simulationstudies and they do not utilize feedback. References 2-6, 2-21–23, and 2-80 describethese types of control. Some approaches do utilize feedback [2-24].

Normal forms. Recent work on nonlinear control using normal forms indicates thatstressed power systems exhibit characteristics that can be addressed by includingadditional terms in the Taylor series expansion of the nonlinear system. References 2-25and 2-26 discuss the basic theory behind normal forms. The standard approach forlinearizing a nonlinear system involves using only the first or linear term in the Taylorseries expansion of the nonlinear system. All higher order terms are neglected in linearanalysis. Normal forms include the effects of some higher order terms in the Taylor seriesexpansion and can provide insight into the modal interactions exhibited by powersystems. Both the linear approach and the normal forms approach use approximations tothe full nonlinear system, but the normal forms approach is able to include more of thesystem nonlinearities. Reference 2-27 is concerned with including second order terms toaffect nonlinear tuning of controller gains.

Dissipativity. Reference 2-52 proposes a unifying framework for analysis and synthesisof controllers to damp low frequency oscillation in power systems. The basic idea is that apassive system always consumes energy. The controllers can be HVDC links, static varcompensators (SVCs), thyristor controlled series capacitors (TCSCs) and power systemstabilizers.

Energy (Lyapunov) function methods. The application of energy (Lyapunov) functionmethods in transient stability analysis of electric power systems is well known [2-71,2-

2-4

72]. In recent years, use of energy function principles to derive control strategies forlarge-scale power systems has received increased research attention [2-73,2-74,2-75].Advantages of energy function control strategies are that the form is independent of thestructure, i.e., structural uncertainty is not a main issue; they may rely on local signals,and they have large regions of validity as they are based on the nonlinear system. A mainlimitation is that the derivation requires that an energy function of the system model befound. This results in modelling assumptions that are rather restrictive. Grönquist et al.[2-75] study the effects of applying controls to FACTS devices based on energy functionmethods for lossless system models.

Reference 2-74 investigates and evaluates transient stability enhancement of large-scalepower systems by control strategies for unified power flow controller, controlled seriescompensation, and phase shifting transformers. The controls are applied to a CIGRÉ testsystem that has dynamic properties similar to the Swedish and interconnected Nordicpower system. Reference 2-73 describes control strategy for HVDC converter controlsbased on energy function methods.

Nonlinear fuzzy and neural net control. As described in Chapter 4, fuzzy system andneural network applications to rotor angle stability problems is a research area. Theadvantage of fuzzy controllers is their ability to incorporate nonlinear effects into theresulting control surfaces. An important problem to overcome in power system anglestability applications is that an expert may not be available to provide guidance informing the fuzzy rules due to the complexity and variability of the dynamic processes.Reference 2-70 describes an integrated fuzzy controller for voltage regulation, powersystem stabilizer and governor control of a generator. Field tests of a fuzzy PSS are alsobriefly described. Neural nets provide another and perhaps complimentary solution to thenonlinear control problem through their capacity to learn from system conditions andmodel nonlinear effects. References 2-28 and 2-29 recent work in this area.

2.2 Linear Control TechniquesPower system linear control design is a process that can be divided into distinct steps; thenumber depends on the situation. One situation arises if the control principle is alreadydecided. Reference 2-30 proposes a three-step design procedure for end-use load control.The steps are: 1) select a location for control actuation, 2) choose feedback signals, and 3)select the compensating parameters.

Another design situation occurs if the control principle is not yet decided. Then theproblem is to find the most cost-efficient way to solve the angle stability problem. Thekey question is to find and evaluate different alternatives. These can range fromengineering work retuning existing controllers such as PSS, to large investments in newpower electronic devices. Some alternatives are listed below:

• Retune existing PSS, AVR, and HVDC link and SVC controllers.

• Upgrade control equipment for existing primary controllers such as HVDC, SVCs.

• Add control equipment to existing devices. For example, load modulation control ofelectrical heaters used in district heating.

2-5

• Add a new power electronic device.

• Strengthen the primary system with a new transmission line.

• Calculate system operating restrictions on-line.

Eigenvalue sensitivity [2-31] and participation factors [2-2] are well-known methods oflocating control equipment; see also references 2-32, 2-33 and 2-62. Structural aspects ofcontrolling active loads are presented in reference 2-32. Reference 2-63 describes acontroller design and analysis approach to adjust the existing structure of a system byretuning the internal control loops to relocate critical zeros, thus removing the constraintsthat arise when zeros are at unsuitable locations. Retuning is based on an existingextension of modal analysis to linear system zeros.

References 2-34, 2-55, and 2-36 discuss the use of transfer function residue informationfor placing and designing controllers. The residue of a transfer function is similar to theparticipation factor of a state space model. Residues provide information about whichmodes are most sensitive to gain variations, and what directions the poles will movewhen the gain is increased.

Modeling and model reduction. Design methods and model reduction are intimatelyrelated and some remarks are appropriate. Many advanced methods, especially for robustcontrol, require extensive computations. Therefore it is not feasible to use design modelsas detailed as those used for time domain simulation. Either we adopt a reduced ordermodel suitable for the design method, or we are restricted to design methods withmoderate computation requirements. In automatic control, it is argued that the best modelis the simplest one that is accurate enough to fulfill the design requirement. It’s importantto find a reasonable compromise between model complexity and the design method’scomputational requirements.

Reference 2-35 describes modeling and model reduction from a control perspective. It’spointed out that model reduction may involve:

a) model order reduction in a linear system;

b) model approximation of a nonlinear differential equations by linear systems;

c) approximation of the nonlinear system by ignoring higher-order harmonics.

Note that the case of model order reduction for high order nonlinear differential equationsto low order nonlinear differential equations is not considered. This is actually thesituation power engineers face when having a complex multi-machine simulation modelthat includes saturation nonlinearities and also nonlinearities in the power flow equations.For case a), MATLAB’s Control System Toolbox offers usable tools for model reduction.References 2-36 and 2-37 present time-scale decomposition applied to power systems.This method is especially suitable for a design aiming at a certain frequency window,such as PSS design. Reference 2-38 outlines how synchrony, a generalization of slow-coherency, can be used to construct dynamic equivalents by aggregation of generators.The method is reported to be effective in decomposing the eigenanalysis ofelectromechanical modes

2-6

Identification of models. Another approach to obtain dynamic models for linearcontroller design in power systems is to use identification techniques on systeminput/output data. In reference 2-55 Prony analysis, modified for transfer functionidentification, is presented and in reference 2-56 this method is extended for robustnessconsiderations. One advantage of this approach is that it can be applied to field data anddoes not strictly rely on simulation studies. The models obtained using these methods aregenerally reduced order because only the system modes observable in the output signalcan be incorporated. Once the transfer function model is obtained any standard linearcontrol design procedure can be used. One must be aware, however, of the limited rangeof validity of these models [2-56]. Additional recent work in this area is reported [2-57,2-58,2-61]. An extension of Prony analysis for multiple output signals is discussed inreferences 2-78 and 2-79.

Robustness. The need for robustness design depends on system properties. What are thepossible operating conditions? Where are the load centers? Where are the generationareas? Are the power flow directions always the same? For example in the Nordel systemthat connects the Scandinavian countries, there is a common market for tradingelectricity. The normal trading pattern gives a power flow from Norway, through theSwedish west coast to Denmark. However, for years with little rain the power flowdirection can be reversed. Here there is an obvious need for a robust design method thatcan handle two very different operating conditions. For other systems, such as the NewSouth Wales system in Australia dominated by coal fired plants and well-defined loadcenters, the need for robust design is less pronounced. Robustness can include manydifferent types of uncertainties and some are listed below.

• Different load flow patterns.

• Varying load levels during the day, week, or year.

• Load characteristics, such as voltage and frequency dependence, that might vary withseasons and time of day. In Sweden, a lot of electric heating is used during winter,and in summer air conditioning can be used. Their voltage and frequency dependenceis very different.

• Uncertainty in the topology (structure) of the power system—some plants, lines ortransformers might be taken out for maintenance.

• The dynamic model of the power system always has some level of parameteruncertainty. Some of these parameters are related to design, such as generator timeconstants and inductance. Once determined, their change is negligible. Otherparameters such as AVR, PSS, and turbine governor are tunable parameters that areeasy to change. Even if these parameters have been identified, there is always a riskof subsequent modification without updating the model.

• Some parameters change slightly during operation. Line resistance is temperaturedependent and Load-Tap-Changers (LTC) can change the nominal transformer ratio.

The control principle itself might be inherently robust, i.e., it works with a very limitedknowledge about the power system. For example direct load switching to damp generator

2-7

oscillations only needs two impedances and one switching level [2-32]. In contrast, thedesign in reference 2-39 is based on a linear multi-machine model of the entire powersystem. Many blackouts are caused by cascading disturbances that were not foreseen.Ultimately the power system should be robust to unforeseen disturbances. Poweroscillations are often triggered by an initial disturbance that can give a range of possibleinput amplitudes or operating conditions to the system. The design should also be robustto variation in disturbance amplitude and operating conditions.

Linear design methods. The linear control design literature is extensive. Many designmethods exist for linear and non-linear systems, and some methods include uncertainty.See references 2-40–43. References 2-2 and 2-44 present overviews of design method forpower system applications. The methods can be categorized in different ways such as:

• Linear (linear output or state feedback) or nonlinear (on-off) control law.

• Linear or nonlinear design method. For example LQ-design can use a nonlinearcriteria to design a linear state feedback.

• By the physical device the design is aiming for, that is, design for PSS, AVR, HVDC,SVC, or load switching.

• By a development scale ranging from academic control methods, to methods used todesign controllers implemented in the power system. The evolution of a designmethod goes through the evolutionary steps: theory, small illustrative simulationstudy, larger simulation study, redesign, preliminary field test, redesign, and finallyworking application in a power system.

It always falls back to engineering judgment when deciding whether an advanced designmethod is really necessary, or if a simple control scheme would be sufficient.Measurements of time synchronized phasors opens new possibilities to feedback lawsthat can be inherently robust. The control design must be simple enough to be reliablyapplied to a physical system.

LQG methodology. Linear quadratic (LQ) control design is an attractive theoreticalapproach that has not found wide application in practice. Reference 2-39 presents a linearquadratic (LQ) based design method used to find a feedback structure and parameters forPSS/AVR. MATLAB software [2-45] is the main modelling and design tool. A linearizedmulti-machine model is used to design an optimal LQ-controller with full state feedback.In LQ design a trade-off is done between input energy and performance. It’s suggestedthat the best generator to damp a certain mode is the one where the optimal controlleruses most of its input energy. Instead of using a full state feedback, the feedback isrestricted to a sparse structure where most signals are local and only a few strategic globalsignals are used. This structure is retuned by parametric LQ, that is, numericalminimisation of the loss criteria used in LQ-design. The method’s strength is that thedesign is done using a multi-machine model, so all PSS and AVRs design is coordinatedand simultaneous. The weak points are that the design is done at one operation point andthe method does not consider robustness. Reference 2-59 provides another example usingLQ design on a very large power system.

2-8

LQG/LTR methodology. Linear quadratic regulators discussed in the previous sectionhave appealing robustness properties, including guaranteed gain margins of 6 dB orgreater and phase margins of at least 60 degrees. However such controllers requireknowledge of all the system states which usually is not possible or practical in powersystem applications. In these cases loop transfer recovery (LTR) can be used to estimateunavailable states and still retain the robustness properties of full state feedback withLQG. LQG/LTR is used to design stabilizing controllers for a SVC in references 2-56 and2-65, and an HVDC link in reference 2-64. Application to power systems is proceeded byidentification of an effective low order transfer function which is used as the designmodel.

Desensitized Control. In reference 2-47 a single-machine infinite bus model is used todesign a robust regulator integrating AVR and PSS functions. In the design the controlleris desensitized, i.e., made insensitive to parametric uncertainties. In this way robustness isincluded in the design. The design method was originally developed for the “Four-Loops-Regulator” structure used by Electricite de France (EdF), but reference 2-46 shows thatthe method can also be used for a standard AVR/PSS structure. The method is used toretune EdF’s voltage regulators, and the new values will soon be used in operation.

Robust Control, µ-design, H∞.. Robust control is a well-established discipline withtextbooks and MATLAB toolboxes [2-43,2-48]. Reference 2-49 proposes a frameworkfor robust stability assessment of controls in multi-machine power systems. StructuredSingular Value (SSV) is used to determine stability for varying operation conditions. Inthe companion paper [2-50], the method is used in a simulation study of a four-machinetest system. The simulation results show excellent accuracy of robust stability assessmentfor a wide range of operating conditions. Reference 2-51 points out that robust controllersdesigned by µ-design can produce extremely fragile controllers in the sense thatvanishing-small perturbations of the coefficients of the designed controller destabilize theclosed-loop control system. Reference 2-60 is another study of H∞ control design topower systems.

Design methods for active load controllers. Control of active load can be used toimprove angle stability. Reference 2-30 describes control of end-user loads in the westernUSA to enhance stability. Reference 2-9 describes modulated controllable loads forpower system stabilization. It’s found that a decentralised two-loop load stabilizer, usinglocal bus voltage and frequency, adds damping to all oscillation modes.

Reference 2-81 presents an on-off damping controller for a single machine system. It wasused during a field test in southern Sweden to damp oscillations at a 0.9 MW hydropower generator. The controller used estimated machine frequency as input andcontrolled a 20 kW resistive load via thyristor switches. The results indicate that on-offcontrol of active loads is effective in terms of added damping, and that it is simple to tuneand implement.

References2-1 P. M. Anderson and A. A. Fouad, Power System Control and Stability, IEEE

Press, revised printing 1994.

2-9

2-2 P. Kundur, Power System Stability and Control, McGraw-Hill, 1994.

2-3 F. P. DeMello and C. Concordia, “Concepts of Synchronous Machine Stability asAffected by Excitation Control,” IEEE Transactions Power Apparatus andSystems, Vol. PAS-88, No 4, April 1969.

2-4 N.B. Bhatt, “Field Experience with Momentary Fast Turbine Valving and OtherSpecial Stability Controls Employed at AEP’s Rockport Plant,” IEEETransactions on Power Systems, Vol. 11, No. 1, pp. 155–161 February 1996.

2-5 Matsuzawa, K. Yanagihashi, J. Tsukita, M. Sato, T. Nakamura, and A Takeuchi,“Stabilizing Control System Preventing Loss of Synchronism from Extension andIts Actual Operating Experience“, IEEE Transactions on Power Systems, Vol. 10,No. 3, pp. 1606–1613, August. 1995.

2-6 C.A. Stigers, C. S. Woods, J. R. Smith and R. D. Setterstrom, “The AccelerationTrend Relay for Generation Stabilization at Colstrip,” IEEE Transactions onPower Delivery, Vol. 12, No. 3, pp. 1074–1081, July 1997.

2-7 H. Ota, Y. Kitayama, H. Ito, N. Fukushima, K. Omata, K. Morita, and Y. Kokai,“Development of Transient Stability Control System (TSC System) Based on On-Line Stability Calculations,” IEEE Transactions on Power Systems, Vol. 11, No.3, pp. 1463–1472, August 1996.

2-8 H. Bourlés, F. Colledani, and M. P. Houry, “Robust Continuous Speed GovernorControl for Small-Signal and Transient Stability,” IEEE Transactions on PowerSystems, Vol. 12, No 1, pp. 129–135, February 1997.

2-9 I. Kamwa, R. Grondin, D. Asber, J. P. Gingras, and G. Trudel, “Active PowerStabilizers for Multimachine Power Systems: Challenges and Prospects,” IEEETransactions on Power Systems, Vol. 13, No 4, pp. 1352–1358, November 1998.

2-10 O. Samuelsson, B. Eliasson, G. Olsson, “Power Oscillation Damping withControlled Active Loads,” paper SPT PS 09-04-0620, Proceedings of theIEEE/KTH Stockholm Power Tech Conference, Stockholm, Sweden, June 10–22,1995, Vol. III, pp. 274–279.

2-11 D. J. Hill (Editor), Special Issue on Nonlinear Phenomena in Power Systems,Proceedings of the IEEE, Vol. 83, no. 11, November 1995.

2-12 L. Gao, L. Chen, Y. Fan, and H. Ma, “DFL-Nonlinear Control Design withApplications in Power Systems,” Automatica, Vol. 28, 1992, pp. 975–979.

2-13 F. Fatehi, J. R. Smith, D. A. Pierre and M. H. Nehrir, “Application of FeedbackLinearization to Generator Speed Control in Multimachine Power Systems,”Proceedings of the 27th Annual North American Power Symposium, BozemanMontana, Oct. 2–3, 1995, pp. 487–492.

2-14 J. R. Smith, D. A. Pierre, I. Sadighi, M. H. Nehrir, and J. F. Hauer, “ASupplementary Adaptive VAR Unit Controller for Power System Damping,”

2-10

IEEE Transactions on Power Systems, Vol. 4, No. 3, pp. 1017–1023, August1989.

2-15 J. Reeve and M. Sultan, “Gain Scheduling Adaptive Control Strategies for HVDCSystems to Accommodate Large Disturbances,” IEEE Transactions on PowerSystems, Vol. 9, No. 1, pp. 366–372, February 1994.

2-16 A. A. Ghandakly and A. M. Farhoud, “A Parametrically Optimized Self-TuningRegulator for Power System Stabilizers,” IEEE Transactions on Power Systems,Vol. 7, No. 3, pp. 1245–1250, August 1992.

2-17 P. Shamsollahi and O. P. Malik, “An Adaptive Power System Stabilizer UsingOn-Line Trained Neural Networks,” IEEE Transactions on Energy Conversion,Vol. 12, No 4, pp. 382–387, December 1997.

2-18 M. Lown, E. Swidenbank, and B. W. Hogg, “Adaptive Fuzzy Logic Control of aTurbine Generator System,” IEEE Transactions on Energy Conversion, Vol. 12,No .4, pp. 394–399, December 1997.

2-19 Y. M. Park, U. C. Moon, and K. Y. Lee, “A Self Organizing Power SystemStabilizer Using Fuzzy Auto-Regressive Moving Average (FARMA) Model,”IEEE Transactions on Energy Conversion, Vol. 11, No 2, pp. 442–448, June1996.

2-20 P. S. Dolan, J. R. Smith and W. A. Mittelstadt, “Study of TCSC OptimalDamping Control Parameters for Different Operating Conditions,” IEEETransactions on Power Systems, Vol. 10, No. 4, pp. 1972–1978, November 1995.

2-21 D. N. Kosterev and W. J. Kolodziej, “Bang-Bang Series Capacitor TransientStability Control,” IEEE Transactions on Power Systems, Vol. 10, No. 2, pp. 915–929, May 1995.

2-22 C. W. Taylor, J. R. Mechenbier, and C. E. Matthews, “Transient ExcitationBoosting at Grand Coulee Third Power Plant: Power System Applications andField Tests,” IEEE Transactions on Power Systems, Vol. 8, No. 3, pp. 1291–1298,August 1993.

2-23 M. L. Shelton, W.A Mittelstadt, P. F. Winkelman, and W. L. Bellerby,“Bonneville Power Administration 1400 MW Braking Resistor,” IEEETransactions on Power Apparatus and Systems, Vol. PAS-94, pp. 602–611,March 1975.

2-24 J. Chang and J. H. Chow, “Time-Optimal Control of Power Systems RequiringMultiple Switchings of Series Capacitors,” IEEE Transactions on Power Systems,Vol. 13, No. 2, pp. 367-373, May 1998.

2-25 S. K. Starret and A. A. Fouad, “Nonlinear Measures of Mode-MachineParticipation,” IEEE Transactions on Power Systems, Vol. 13, No. 2, pp. 389–394, May 1998.

2-11

2-26 G. Jang, V. Vittal, and W. Kliemann, “Effects of Nonlinear Modal Interactions onControl Performance: Use of Normal Forms Technique in Control Design: Part I:General Theory and Procedure,” IEEE Transactions on Power Systems, Vol. 13,No. 2, pp. 401–407, May 1998.

2-27 G. Jang, V. Vittal, and W. Kliemann, “Effects of Nonlinear Modal Interactions OnControl Performance: Use of Normal Forms Technique in Control Design: Part II:Case Studies,” IEEE Transactions on Power Systems, Vol. 13, No. 2, pp. 408–413, May 1998.

2-28 K. A. El-Metwally, G. C. Hancock, and O. P. Malik, “Implementation of a FuzzyLogic PSS Using a Micro-Controller and Experimental Test Results,” IEEETransactions on Energy Conversion, Vol. 11, No. 1, pp. 91–96, March 1996.

2-29 M. A. Abido and Y. L. Abdel-Magid, “A Hybrid Neuro-Fuzzy Power SystemStabilizer for Multimachine Power Systems,” IEEE Transactions on PowerSystems, Vol. 13, No 4, pp. 1323-1330, November 1998.

2-30 J. E. Dagle, D. W. Winiarski, and M. K. Donnelly, End-Use Load Control forPower System Dynamic Stability Enhancement, Pacific Northwest Laboratory,Report PNNL-11488, prepared for the U.S. Department of Energy under ContractDE-AC02-76RLO 1830, February 1997.

2-31 B. Eliasson, Damping of Power Oscillations in Large Power Systems, Ph.D.-thesis, CODEN: LUTFD2/(TFRT-1032)/1-155/(1990), Department of AutomaticControl, Lund Institute of Technology, Lund, Sweden, 1990.

2-32 O. Samuelsson, Power System Damping—Structural Aspects of ControllingActive Power, Ph.D.-thesis, CODEN: LUTEDX/(TEIE-1014)/1-196/(1997),Department of Industrial Electrical Engineering and Automation, Lund Institute ofTechnology, Lund, Sweden, 1997.

2-33 N. Yang, J. D. McCalley, and Q. Liu, “TCSC Controller Design for DampingInterarea Oscillations,” IEEE Transactions on Power Systems, Vol. 13, No 4, pp.1304–1310, November 1998.

2-34 M. E. Aboul-Ela, A. A. Sallam, J. D. McCalley and A. A. Fouad, “ DampingController Design for Power System Oscillations Using Global Signals,” IEEETransactions on Power Systems, Vol. 11, No. 2, pp. 767–773, May 1996.

2-35 R. Johansson, System Modeling and Identification, Prentice-Hall, 1993.

2-36 J. H. Chow, Time-Scale Modeling of Dynamic Networks with Application toPower Systems, Lecture Notes in Control and Information Sciences, Vol. 46,Springer-Verlag, 1982.

2-37 P. W. Sauer, D. J. LaGesse, S. Ahmed-Zaid, and M. A. Pai, “Reduced OrderModeling of Interconnected Multimachine Power Systems Using Time-ScaleDecomposition,” IEEE Transactions on Power Systems, Vol. PWRS-2, No. 2, pp.310–320, May 1987.

2-12

2-38 G. N. Ramaswamy, G. C. Verghese, L. Rouco, C. Vialas, and C. L. DeMarco,“Synchrony, Aggregation and Multi-Area Eigenanalysis”, IEEE Transactions onPower Systems, Vol. 10, No. 4, pp. 1986–1993, November 1995.

2-39 M. Akke, Power System Stabilizers in Multimachine Systems, Tech. Lic.-thesis,CODEN: LUTFD2/(TFRT-3201)/1-103/(1989), Department of AutomaticControl, Lund Institute of Technology, Lund, Sweden, 1989.

2-40 B. D. O. Anderson, and J. B. Moore, Optimal Control: Linear QuadraticMethods, Prentice-Hall, Englewoods Cliffs, N. J., USA, 1989.

2-41 J. M. Maciejowski, Multivariable Feedback Design, Addison Wesley,Wookingham, UK, 1989.

2-42 J. J. E. Slotine and W. Li, Applied Nonlinear Control, Prentice-Hall, 1991.

2-43 K. Zhou, J. C. Doyle, K. and Glover, Robust Optimal Control, Prentice-Hall,1995.

2-44 CIGRÉ TF 38.01.07, Analysis and Control of Power System Oscillations,Brochure 111, December 1996.

2-45 MATLAB—User’s Guide, The MathWorks, Inc., Natick, MA, USA, 1992.

2-46 H. Bourlés, S. Peres, T. Margotin, M. P. Houry, “Analysis and Design of a RobustCoordinated AVR/PSS”, IEEE Transactions on Power Systems, Vol. 13, No 2,pp. 568–575, May 1998.

2-47 A. Heniche, H. Bourlés, and M. P. Houry, “A Desensitized Controller for VoltageRegulation of Power Systems,” IEEE Transactions on Power Systems, Vol. 10,No. 3, pp. 1461–1466, August, 1995.

2-48 G. J. Balas, J. C. Doyle, K. Glover, A. Packard, and R. Smith, µ-Analysis andSynthesis ToolBox, Mathworks, Natick, MA, 1993.

2-49 M. Djukanovic, M. Khammash, V. Vittal, “Application of the Structured SingularValue Theory for Robust Stability and Control Analysis in Multimachine PowerSystems—Part-I: Framework Development,” IEEE Transactions on PowerSystems, Vol. 13, No 4, pp. 1311–1316, November 1998.

2-50 M. Djukanovic, M. Khammash, V. Vittal, “Application of the Structured SingularValue Theory for Robust Stability and Control Analysis in Multimachine PowerSystems—Part-II: Numerical Simulation and Results,” IEEE Transactions onPower Systems, Vol. 13, No 4, pp. 1317–1322, November 1998.

2-51 L. H. Keel and S. P. Bhattacharyya, “Robust, Fragile, or Optimal?,” IEEETransactions on Automatic Control, Vol. 42, No. 8, August 1997.

2-52 A. M. Stankovic, P. C. Stefanov, G. Tadmor, And D. J. Sobajic, “Dissipativity asa Unifying Control Design Framework for Suppression of Low FrequencyOscillations in Power Systems,” IEEE Transactions on Power Systems, Vol. 14,No 1, pp. 192–199, February 1999.

2-13

2-53 M. Klein, G. J. Rogers, and P. Kundur, “A Fundamental Study of InterareaOscillations in Power Systems,” IEEE Transactions on Power Systems, Vol. 6,No. 3, pp. 914–921, August 1991.

2-54 J. W. Chapman, M. D. Illic, C. A. King, L. Eng, and H. Kaufman, “Stabilizing aMultimachine Power System via Decentralized Feedback Linearizing ExcitationControl,” IEEE Transactions on Power Systems, Vol. 8, No. 3, pp. 830–839,August 1993.

2-55 J. R. Smith, F. Fatehi, C. S. Woods, J. F. Hauer and D. J. Trudnowski, “TransferFunction Identification in Power System Applications,” IEEE Transactions onPower Systems, Vol. 8, No. 3, pp. 1282–1290, August 1993.

2-56 F. Fatehi, J. R. Smith and D. A. Pierre, “Robust Power System Controller DesignBased on Measured Models,” IEEE Transactions on Power Systems, Vol. 11, No.2, pp. 774–780, May 1996.

2-57 A. B. Leirbukt, J. H. Chow, J. J. Sanchez-Gasca, and E. V. Larsen, “DampingControl Design Based on Time-Domain Identified Models,” IEEE Transactionson Power Systems, Vol. 14, No 1, pp. 172–178, February 1999.

2-58 J. J. Sanchez-Gasca and J. H. Chow, “Computation of Power System Low-OrderModels from Time Domain Simulations Using a Hankel Matrix,” IEEETransactions on Power Systems, Vol .12, No. 4, pp. 1461–1467, November 1997.

2-59 A. J. A. Simoes Costa, F. D. Freitas, and A. S. de Silva, “Design of DecentralizedControllers for Large Power Systems Considering Sparsity,” IEEE Transactionson Power Systems, Vol. 12, No. 1, pp. 144–152, February 1997.

2-60 G. N. Taranto and J. H. Chow, “A Robust Frequency Domain OptimizationTechnique for Tuning Series Compensation Damping Controllers,” IEEETransactions on Power Systems, Vol. 10, No. 3, pp. 1219–1225, August. 1995.

2-61 I. Kamwa, G. Trudel, L. Gerin-Lajoie, “Low-Order Black-Box Models for ControlSystem Design in Large Power Systems,” IEEE Transactions on Power Systems,Vol 11, No. 1, pp. 303–312, February 1996.

2-62 J. H. Wilkinson, The Algebraic Eigenvalue Problem, Oxford University Press,London, 1965.

2-63 L. Jones and G. Andersson, “Application of Modal Analysis of Zeros to PowerSystem Control and Stability,” Electric Power Systems Research, Vol. 46, No. 3,pp. 205–211, September 1998.

2-64 L.E. Jones, H. Bourles and G. Andersson, “Nonconventional Control of HVDCUsing LQG/LTR Methodology,” IEEE Power Engineering Review, 19(1) pp. 62–64, January 1999.

2-65 J. R. Smith, F. Fatehi, and D. A. Pierre, “Closed-Loop Identification and Tuningfor Damping Interarea Modes,” Proceedings of the 33rd Conference on Decisionand Control, pp. 4055–4060, December 1994.

2-14

2-66 S. Kaprielian, K. Clements, and J. Turi, “Feedback stabilization for an AC/DCPower System Model,” Proceedings: 29th Conference on Decision and Control,pp. 3367–3372, December 1990.

2-67 D. J. Trudnowski, and J. E. Dagle, “Effects of Generator and Static LoadNonlinearities on Electromechanical Oscillations,” IEEE Transactions on PowerSystems, Vol.12, No. 3, pp. 1283–1289, August 1997.

2-68 E. V. Larsen, J. J. Sanchez-Gasca and J. H. Chow, “Concepts for Design ofFACTS Controllers to Damp Power Swings,” IEEE Transactions on PowerSystems, Vol. 10, No. 2, pp. 948–955, May 1995.

2-69 E. V. Larsen and J. H. Chow, “SVC Control Design Concepts for SystemDynamic Performance,” IEEE Special Publication # 87TH0187-5-PWR,Application of SVC for System Dynamic Performance, 1987.

2-70 T. Hiyama, Y. Ueki, and H. Andou, “Integrated Fuzzy Logic Generator Controllerfor Stability Enhancement,” IEEE Transactions on Energy Conversion, Vol. 12,No. 4, December 1997.

2-71 D. Hill, and C. N. Chong, “Lyapunov Functions of Lure’-Postnikov form forStructure Preserving Models of Power Systems,” Automatica, Vol. 25. No. 3, pp.453–460, 1989.

2-72 M. A. Pai, Energy Function Analysis for Power System Stability, KluwerAcademic Publishers, 1989.

2-73 K. R. Pai and H. S. Y. Sastry, “A Structure Preserving Energy Function forStability Analysis of AC/DC Systems,” in Sadhana, Vol. 18, Part 5, pp. 787–799,1993.

2-74 M. Ghandari, Control of Power Oscillations in Transmission Systems UsingControllable Series Devices, Licentiate thesis, TRITA-EES-9705, Electric PowerSystems, Royal Institute of Technology, 1997.

2-75 J. J. Grönquist, W. A. Sethares, F. L. Alvarado, and R. H. Lasseter, “PowerOscillation Damping Control Strategies for FACTS Devices using LocallyMeasurable Quantities,” IEEE Transactions on Power Systems, Vol. 10, No. 3,pp. 1598–1605, August 1995.

2-76 L. E. Jones, and G. Andersson, “Robust Controllers for Power Systems: Open andClosed Loop Design Approaches,” Paper No. 21, presented at Electric PowerSystems Operation and Management Conference, ETH, Zürich, September 23–25,1998.

2-77 M. Klein, L.X. Lee, G. J. Rogers, S. Farrokhpay, and N. J. Balu, “H∞ DampingController Design in Large Power Systems,” IEEE Transactions on PowerSystems, Vol. 10, No. 1, pp.158–166, February 1995.

2-15

2-78 D. J. Trudnowski, J. M. Johnson, and J. F. Hauer, “SIMO System Identificationfrom Measured Ringdowns,” Proceedings of the American Control Conference,pp. 2968–2972, June 1998.

2-79 D. J. Trudnowski, J. M. Johnson, and J. F. Hauer, “Making Prony Analysis MoreAccurate Using Multiple Signals,” IEEE Transactions on Power Systems, Vol. 14,No 1, pp. 226–231, February 1999.

2-80 D. C. Lee and P. Kundur, “Advanced Excitation Controls for Power SystemStability Enhancement,” CIGRÉ, paper 38-01, 1986.

2-81 O. Samuelsson, and M. Akke, “On-Off Control of an Active Load for PowerSystem Damping-Theory and Field Test,” IEEE Transactions on Power Systems,Vol. 14, No. 2, pp. 608–613, May 1999.

Chapter 3

State-of-the-Art in Digital Control

The evolution of microprocessor technology—with the consequent availability of reliable,high-performance, low-cost digital hardware and powerful software tools, along with thegrowing difficulties in the maintenance of analog apparatus—has resulted in developmentof powerful digital control systems. The progress in microprocessor technology has led tocontinuously increasing performance in term of speed, computing power andfunctionality, through the integration on one chip of functions that in the past requiredmany external components.

A similar evolution has taken place also in the software field: very powerful developmentenvironments have been carried out by specialized software houses and offered on themarket. Such environments include a wide set of tools (debugger, software analyzer,profiler, I/O, graphic and mathematical libraries, etc.) for making the code developmentfast and easy, and for allowing its independence from the hardware. The availability ofreliable hardware and software products together with their high performance and theassociated low costs makes the use of the digital technology convenient and feasible formost power system controls. The benefits from the use of digital technology include:

• Greater flexibility and adaptability to different practical needs;

• Reduced number of subset types (electronic boards) to be used for the practicalrealization;

• Improvement in the user interface which becomes graphic, friendly and interactive;

• Enhanced control, alarm and protection functions;

• Sophisticated auto-diagnostics on-board;

• Easy and accurate setting and change of control parameter values, their constancy andindependence from environmental conditions.

Moreover, the microprocessor technology makes easy the addition of new functionalitysuch as:

• Adaptive or non-linear control;

• Data communication with the supervisory systems;

• On-line monitoring and the transient recording of meaningful control and processvariables;

• Simulation of the unit under operation for a check of control parameter values withoutinterfering with the plant.

Table 3-1 compares analog and digital control.

3-2

Table 3-1: Comparison between control types.

Item Analog control Digital controlProcessing Each operation is synthesized by a

physical device (sums, integration,limitations, etc.). New functionsrequire the addition of new hardware.

Functions are synthesized in software. No extrahardware is required for implementing newfunctions.

Modelidentification

Very expensive, causes machineunavailability. Accuracy depends onsignal measuring quality

Direct translation of the digital control model fromthe studies and simulation platform.

Testing Very expensive. Difficult execution.A lot of instrumentation and otherfacilities are required.

Many testing facilities can be made available,considerably reducing costs and risks. Data-acquisition automatically provides recording ofmany physical and internal variables; Softwareemulates instrumentation for different kinds oftesting, such as step and sinusoidal signals, etc.;

Commission-ing

The need for a lot of instrumentationand numerous calibration tasks makethe commissioning a quite strenuousand time-consuming job.

Commissioning can be carried on easily and in ashorter time because parameters are expressed inper unit or seconds, according to the mathematicalmodel specified. A computer-based simulationstation can be used for pre-commissioning andtraining. Adjustments can be automaticallydocumented. Facilities such as recording, settingsand readings make performance verification veryeasy.

Resolution Theoretically infinite. In practicelimited due to noise, drift, etc..

Limited to the converters and to the length of theword used to make the calculations. At the presenttime 16-bit A/D converters are used and, in someapplications, 24-bit are available, to get maximumaccuracy, although with a compromise for speed.Calculations are done in floating point, and with theavailable 80-bit co-processors, accuracy ispractically unlimited..

Stability ofParametervalues

Requires the use of very expensivecomponents in every circuit board.Strongly affected by componentaging. Thermal drift causesparameter limitations. Maintenanceservices produce parameter changes.

Digital implementation provides drift-free settings.Only analog interfaces need expensive and low driftcomponents. Analog time constants used in theseinterfaces are of little importance to the processparameters. Parameters don’t change because ofmaintenance services.

Interfaces Done through conventional switchesand instruments.

Interface design using human factors engineeringconcepts such as: familiar and standardnomenclature, engineering units, facilities forunderstanding and performing critical tasks;usability (dialogs, messages, acknowledgment,graphics).

Control Laws Practically restricted to linearapplications, with severe constraintsfor complex laws. Adaptive controllaws are difficult to implement,requiring an excessive number ofsimplifications.

Complex control laws, such as adaptive control, canbe easily implemented. Fuzzy controllers areequally feasible. Facilities to test and perform newideas.

3-3

Item Analog control Digital controlSoftware None The real time core manages the execution of the

various tasks. Software for off-line analysis areavailable.

Connectivity The connection with the other plantdevices is done through relays andanalog signals.

Focus is held on serial interfaces - such as RS485 -with protocols such as ModBus. The growing use ofEthernet TCP/IP as field bus (suitable cables andconnectors are being developed) allows direct andremote connection with the analyst even viaInternet. Communication resources exist for remotecontrol and integration with supervisory systems forpurposes such as: remote settings changing, remotedata logging, and customer protocols compatibility.

Self-diagnosis None. Wide self-diagnosis resources. New techniques arebeing developed based upon artificial intelligenceconcepts.

Costs Stable and relatively low. Decreasing. cost/benefit relationship may still beunfavorable in small size power plants.

Maintenance Expensive due to the lack of self-diagnosis.

Can be considerably cheaper because of the use ofthe self-diagnosis. A trend towards not repairingprinted circuit boards is being observed.

Bandwidth Can be very large. Still limited for extremely fast and complex loopsbecause the Real Time Operating Systems need tokeep switching between multiple tasks. For goodperformance, the times between the task switchingand the latency for the interruptions must be smallerthan the smallest control sampling rate. Use ofassembly instructions to accelerate the process canbring serious maintenance problems.

This chapter examines the following topics:

• Section 1 reviews the fundamentals of digital control of dynamic systems.

• Section 2 describes the basic structure of a digital-control system.

• Section 3 describes application of digital control for a generator excitation system.

• Section 4 describes application of digital control for static var compensators.

3.1 Review of Digital Control of Dynamic SystemsFigure 3-1 shows a computer-controlled dynamic system.

The output from the process )(ty is a continuous-time signal. The output is converted

into digital form { })( kty by the analog-to-digital (A-D) converter. The conversion is done

at the sampling times, kt . The computer processes the measurements using an algorithm,

and gives a new sequence of numbers { })( ktu . This sequence is converted to an analog

signal by a digital-to-analog (D-A) converter. The real-time clock in the computersynchronizes the events.

3-4

A-D D-A

ClockComputer

Algorithm Process)( kty )( ktu )( ktu )( kty

Fig. 3-1. Schematic diagram of a computer-controlled system.

3.1.1 Sampling of continuous-time signalsAssume that the continuous-time system is given in the following state-space form:

)()()(

)()()(

tuDtxCty

tuBtxAtx

+=+=&

The system has r inputs, p outputs, and is of order n. Normally a D-A converter isconstructed so that it holds the analog signal constant until a new conversion is ordered.The relationship between the system variables at the sampling instants can be determined.Given the state at the sampling time kt the state at the next sampling time 1+kt is thus

given by:

( ) ( )∫+

++ ′′+= ′−−+

1

11 )()()( 1

k

k

kkk

t

t

stAk

ttAk sdsBuetxetx

The system equation of the sampled system is:

( ) ( ))()()(

)(,)(,)( 111

kkk

kkkkkkk

tuDtxCty

tutttxtttx

+=Γ+Φ= +++

where:

( )

( ) ∫−

+

−+

+

+

=Γ

=Φkk

kk

ttAs

kk

ttAkk

Bdsett

ett

1

1

01

)(1

,

,

Example 3-1. Consider the first order system:

uxtx βα +=)(&

For periodic sampling with periodτ ,

τktk =

3-5

Applying the formulas above we get:

( )10

−==Γ

=Φ

∫ αττ

α

ατ

αββ edse

e

s

The samples system thus becomes:

( ) ( ) )(1)( ταβτττ ατατ kuekxekx −+=+

Pulse transfer operator. Use of the pulse-transfer operator allows the input-outputrelationship to be conveniently expressed as a rational function )()()( kuqHky = where:

DqCqH +ΓΦ−Ι= −1)()(

where q is a shift operator with

)1()( += kxkxq

Example 3-2. For a second-order single-input, single-output we have:

22

11

22

110

1

1

)(

)(

)()(

−−

−−

−

++++

=

=

+ΓΦ−Ι=

qaqa

qbqbb

qA

qB

DqCqH

This means that the input-output model can be written as:

)2()1()()2()1()( 21021 −+−+=−+−+ kubkubkubkyakyaky

Digital filtering. Digital filtering provides a great deal of flexibility, since the filtercharacteristic can easily be changed by tuning a few parameters. A digital filter has thegeneral form:

)(...)1()()(...)2()1()( 1021 mkubkubkubnkyakyakyaky mn −++−++−−−−−−−=

where y is the filter output and u is the input measurement value. If all the a parametersare zero we will have a moving average filter with a finite impulse response. If some orall of a parameters are non-zero there is an auto-regressive filter which has an infiniteimpulse response. As a numerical example, a second order low-pass filter with a cutofffrequency of 300 Hz can be modeled by:

)2(0278.0)1(0557.0)(0278.0)2(4752.1)1()( −+−++−−−−= kukukukykyky

Poles and zeros. The poles of a system are the zeros of the denominator of )(qH or theeigenvalues of Φ . Because )exp( τA=Φ it follows from the properties of matrixfunction that

3-6

τλλ )()( Ai

ie=Φ

The equation above gives the mapping from the continuous-time poles to the discrete-time poles. Through this analysis it’s obvious that the left half of the s-plane is mappedinto the unit disc of the z plane.

3.1.2 Dynamic performanceFor real-time digital control, the criteria and algorithms for numerical integration ofdifferential equations must result in numerical solutions close to the solutions of thecorresponding continuous-time equations.

The basic point to be deeply considered is the altered dynamics of the system undercontrol when moving from the theoretical description by continuous-time differentialequations to the practical implementation where finite-differences algebraic equations(discrete-time dynamic system) are used [3-9].

This aspect is very important for real-time applications. Because of computing timeconstraints, it’s not always possible to use complex numerical integration algorithmscombined with very short integration step length ( )τ . Thus, the correspondence of digitalcontrol to the nominal analog performance must be verified.

In the following, the adequacy of several numerical integration methods are considered interms of altered poles and residues of the related system transfer functions.

Computation of the altered dynamics. According to the above and the results shown inAppendix B, the critical factor affecting the dynamic behavior of digital control systemsare the numerical integration method and the integration step length.

Analyzing the dynamic behavior of discrete systems it should be guaranteed that the“spurious modes” due to the integration algorithm are stable and timely convergent, andalso that the variations ∆λ, ∆c of initial eigenvalues λ and of the related residue c arenegligible. The value of the integration step τ strongly affects the highest poles of thediscrete-time system: the lower the value, the better the discrete time system approachesthe corresponding continuous model, but the higher the digital hardware performancerequirement. For small variation (sensitivity method) of original generic pole (∆λ/∆ «1),the following relations allow evaluation of the corresponding altered dynamics:

where

r = λτ

is the “equivalent integrator” of the numerical integration method (see Appendix B).

1)( −≅∆rQ

λλ

:/)( ssQ τ

3-7

In case of Explicit Euler (EE) integration method:

An exact calculation of the dynamic performance modification is possible for single stepintegration methods. In fact, coming back to the first order linear differential equation:

that allows a complete and correct analysis of a given mode (λ) of a linear “diagonal”dynamic system (see also the conclusion of the first section of Appendix B), the solutionof this equation can be expressed as the sum of a particular integral I(t), selectedaccording to u(t) and a term proportional to eλt:

x(t)=aeλt +I(t), (a being a suitable constant).

By applying a single-step integration method, the differential equation becomes a finite-difference equations of type:

kkK xx αλτρ +=+ ),(1

where ρ(τ,λ) is a function whose structure depends on the integration method, and αk is,in the most general case, a linear combination of the input u(t) and its derivatives,calculated at instant tk and tk+1 (and possibly at instants between tk and tk+1).

The theory of linear finite-difference equations shows the solution of the above equationto be sum of a “particular integral” Ik , selected knowing the following values of αk and aterm proportional to ρk, where ρ is the root of the characteristic polynomial associated tothe difference equation:

The xk values can be considered as the values, calculated for t=tk=kτ, i.e. sample-values,of a continuous-time function )(tx

)given by:

)()( tIeatx t))) )

+= λ

assuming

kIkI))

)

=

=

)(

ln1

τ

ρτ

λ

We can see that the solution of the discrete system is formally similar to that of the

continuous system. In particular, it’s evident that λ)

affects the solution of the discretesystem like λ affects the solution of the continuous system.

r

rdQr

c

c )(+∆≅∆λλ

)1()(

−=

re

rrQ

u(t)txtutxfdt

tdx +== )( )]();([)( λ

constant) suitablea being ( aIax kk

k

∩+= ρ

3-8

λ)

can therefore be taken as the eigenvalue of the discrete system and can be comparedwith the corresponding eigenvalue of the continuous system.

In this regard, let λλλ ∆+=)

; thus we have:

[ ]1

)(ln −=∆r

rρλλ

This equation highlights the exact transformation that, depending on the specific single-step integration method, links the continuous system eigenvalue to the discrete systemeigenvalue.

Table 3-2 shows the results obtained for different integration algorithms. As an example,with EE method and τ = 10 ms (integration step), the change of a 20 ms continuousmodel time constant is: ∆λ = λ/4. Then, the resulting modified time constant of thecorresponding discrete model is 15 ms, instead of 20 ms.

The table also shows, considering an oscillating system (imaginary eigenvalues), that theEE integration algorithm results in instability.

3.2 Basic Structure of a Digital Control SystemsThe general structure of a process computer interacting with a physical process consistsof the following parts:

• central data processing unit

• process communication channels

• A/D and D/A converters

• sensors and actuators

• physical process.

Figure 3-2 shows the basic structure of a control system:

The physical process is observed with sensors. Conversely, the process is influencedthrough actuators. A digital control system works only on information in numerical form,therefore the collected electrical variables have to be converted via analog to digital(A/D) converters. Information from different source points distributed in space is broughtto the central unit via communication channels. The central control unit interprets allincoming data from the physical process, sends control signals, take decisions on thebasis of the program instructions, exchanges data with the human operators and acceptstheir commands.

3-9

Table 3-2: Altered dynamics due to numerical integration method.

Integration

Method

∆λ/λ ∆λ Pole Shift

EE − +1

2τλ ... − +1

22τλ ...

Im

∆λ

∆λ Re

EUTRAP ( )− +1

62τλ .... − +1

62 3τ λ ...

Im

∆λ

∆λ Re

Runge-Kutta 3

( )− +1

243τλ .... − +1

243 4τ λ ...

Im

∆λ

∆λ Re

Runge-Kutta 4

( )− +1

1204τλ .. − +1

1204 5τ λ ...

Im

∆λ

∆λ Re

EXTRA ( )− +5

122τλ .... − +5

122 3τ λ ...

Im

∆λ

∆λ Re

Computer structures. A computer system is normally built around a central processingunit (CPU) to which are connected peripheral units performing different functions:keyboard, video interface, disk driver, input/output (I/O) cards. Figure 3-3 shows theconventional organization of a computer system. In this configuration, the peripheralunits may communicate directly only with the CPU and only one peripheral unit at thetime may be active exchanging data.

The CPU-centered configuration is inherently inefficient because all data has to passthrough the CPU, even if the CPU does not need it. It’s more effective to design acomputer system where the peripheral units are more independent and have addedcomputing capacity. The peripheral units are connected together with a bus by which eachunit can communicate with all the others. Figure 3-4 shows the principle of a bus-organized computer system.

3-10

Fig. 3-2. The basic structure of a digital-control system

CPU

Disk I/O card

Tape Clock

Printer

Terminal

RAM memory

Fig. 3-3. The conventional organization of computer systems.

In the next two sections we examine two applications of digital control in power systems.

3-11

CPUDisk

I/O card

Tape

Bus

Clock

Printer Terminal

RAM memory

Fig. 3-4. Principle of bus organization.

3.3 Evolution of Excitation Control Systems through MicroprocessorTechnology

The general scheme of a modern static excitation system is shown in Figure 3-5. Itconsists of two parts, respectively named control unit and power unit.

MeasurementsTransducers

AutomaticRegulator

AlarmsProtections

OperatorInterface

Adaptive orNon-Linear Control

Communication

Three-phase thyristorBridge

Gate-PulseControl

On-LineMonitoring

Control Logic

Fig. 3-5. Principle scheme of a modern static excitation system.

In the control unit the blocks marked with solid lines represent the conventional functionssuch as measurement of process quantities, control logic, automatic regulation, alarmsand protection, and phase control of firing pulse. The blocks with dashed line show theadditional functions that can be introduced using a digital control system. The power unitsupplies the excitation current to the field winding of the generator and mainly consists ofa three-phase full-controlled thyristor bridge.

In the following a short description of the characteristics and performances of theconventional analog control unit is given, as reference for the requirement specificationand the correct design of a digital one. Referring to Figure 3-5:

• The first block on the left includes circuits for measurement and computation of thefollowing process quantities: active and reactive power, electrical machine voltageand flux, HV bus voltage, excitation voltage and current, generator frequency or

3-12

speed. The major part of these measurements requires high resolution (about 12 bitfor digital transducers) and fast response (response time less than 20 ms).

• The second block Automatic Regulator consists of several control loops. The mainloop regulates the stator voltage and has additional feedback for improving theelectromechanical stability (power system stabilizers, PSS) and for compensating thereactive power drop (compounding). Auxiliary loops limit the working point ofgenerator in over/under excitation and the maximum stator flux. A further possibleauxiliary loop, overlapping the previous, regulates the machine reactive power. Themain loop requires a bandwidth of 5–10 radians/second.

• The block on the right controls the phase of thyristor firing pulses. It maintains thefiring angle inside the allowed range, compensates the gain variations (depending onsupply voltage) and makes the bridge transfer characteristic linear.

• The remaining two marked blocks represent the control and the protection logic. Theymanage the different operating modes of AVR, detect fault or incorrect operatingconditions, and provide proper alarm signals in order to improve the safety and thereliability of the system.

Passing from the analog to the microprocessor technology, the most critical problems,requiring particular care in the design phase are:

• The accuracy, resolution and time response of measurement, transducers, and thyristorfiring pulse phase modulation;

• The dynamic performance of control loops taking into account the altered dynamicsfrom the sample and hold of the I/O signals;

• The reliability and availability of the practical realizations.

3.3.1 Hardware architectureThe hardware configuration of the newest digital AVRs with decentralized architectureconsists of a central system and of modular terminal boards which are placed close to themeasurement points. The central system (see Figure 3-6) mainly consists of CPU andA/D-D/A conversion boards, which communicate via a local bus. It performs measuring,filtering, regulation, logic and communication tasks as well as sampling and holding ofthe measurements and control variables requiring fast management. Less critical data aremanaged by the modular terminal boards which achieve a distributed I/O. Theseperipheral boards communicate with the central system by a field bus.

3-13

Exciter + Generator

Modular Terminal Board nModular Terminal Board 1

Field Bus

Local Bus

Central system

Fig. 3-6. AVR typical hardware architecture.

3.3.2 Software organization and development environmentThe AVR software, executed by the central system CPU boards, implements measuring-filtering, regulation, logic, firing, monitoring and communication functions. The softwareis normally organized in tasks, characterized by different execution frequencies, in orderto optimize the hardware resources and achieve the required dynamic performances. Forexample filtering functions have high execution frequency to avoid possible altereddynamics of the fast control loops, whereas the communication with the human-machineinterface can be executed with lower priority without decreasing the overall performance.To manage the CPU time and the task execution, “ad hoc” schedulers, real-time kernelsor operating systems can be adopted. They give the CPU control to the tasks with theright frequency according to priority level. “Ad hoc” schedulers usually lead to optimized,but quite rigid solutions; a few software changes can require a new plan of the scheduler.Operating systems are more flexible, allowing a plain and structured solution. They canalso manage complex resources, for example drivers and protocols for the communicationwith Local Area Networks (LAN).

Another advantage of operating systems is portability, while the disadvantage is the non-optimized use of resources and therefore the requirement of more powerful digitalhardware. Real-time kernels are an intermediate solution; in spite of lower use of CPUresources they are able to supply useful primitives to organize the software execution.

Similar considerations characterize the software organization: low-level programminglanguages lead to optimized solutions, whereas high-level programming languages andstructured programs lead to plain solutions. General purpose real-time developmentenvironments are present on the market today, providing useful tools to write, to compileand to test the software. Some are PC-based. They permit remote debugging bydownloading the machine code to the CPU boards memory and by monitoring theexecution as normal debuggers, showing the high or low level code processing.

3-14

3.3.3 Reliability and safetyCare in the hardware choice concerns the possible reliability improvement for embeddedreal-time systems. Digital hardware able to run code resident on EPROM or flash-EPROM and to store operative parameters on EEPROM is employed. Watchdog circuitsare usually required on the boards to detect CPU crash and other fatal conditions. Thesoftware can also improve the reliability. It verifies the measurement coherence, filtersthe digital inputs and monitors the correct work of peripherals. Other precautions, forexample cyclic redundancy checks and check-sums, are used to control and to recoverdata errors. Diversified AVR configurations are used for different plant sizes. The AVRsof the largest generators have two and sometimes three central controls. The most popularredundancy configuration is two identical channels: one is active while the other isstandby, waiting to become active if the first malfunctions.

3.3.4 Operator interfaceA friendly and effective operator interface can easily be implemented. It allows both easyand accurate setting of customized data and regulation parameters, and of the on-linedisplay of the most important process variables. Different Human-Machine Interfaces(HMI) are possible, going from a simple LCD display and dedicated push-button to acolored graphical monitor with standard or dedicated keyboard up to a portable PC. Dataexchange with the control room supervisory system is possible through a LAN.

3.4 Application of Digital Control to SVCsAs an example from one manufacturer, Figure 3-7 shows an overview of the ABBMACH 2 computer structure and indicates how the control system interfaces with thehigh voltage components of a SVC.

3.4.1 Control and protection designThe control and protection system provides the following features:

• control functions

• valve control

• protection

• alarms

• operator interface (locally and remotely)

• transient fault recorder

• internal supervision

• remote interrogation

3-15

Main Circuit

Delivery Limit

Control & Protection PC3-phase voltage measurements

Breaker Interface

3-phase current measurements

Digital input

Digital output

110 V

Communication with MACH2 PC

Industrial PC

Control & ProtectionLocal Control Room

Pentiumprocessor

Ethernet boardTCP/IP

PC Motherboard

PCIbus

PC LAN

I/O rack

AC VoltageMeas.

PS 841

AC CurrentMeas.

PS 845

Power SupplyPS 890

BusConnection

PS 870

SwitchControlPS 850

AnalogI/O

PS 860

Isol AnalogInput

PS 862

DigitalInput

PS 851

DigitalOutputPS 853

BackplanePS 880

Supervision BoardPS820

Digital Signal ProcessorBoard PS 801

MultiscanTrinitro n

OWSGWS

COMPAQDESKPRO

BusConnection

PS 930

EVT

Modem

VCU

Fig. 3-7. MACH 2 computer structure.

All functions within the control and protection system are realized with the MACH 2building blocks, consisting of a main computer and an I/O system. High speedapplications, e.g., fast regulators, firing control etc. are realized in DSPs, while lessdemanding functions, such as operators interface, are realized in the main computer CPU.

The I/O rack serves as an intelligent interface between the main computer and the highvoltage side equipment. It contains digital and analog I/O boards, and also a DSP board,PS860. The communication between the main computer and the I/O is by field busses.

3.4.2 CommunicationLocal Area Networks (LANs) are used to connect together several locations (callednodes) so that they can all communicate with each other. The LAN used is the well-known IEEE 802.3 (Ethernet). This bus uses the well-proven Carrier Sense MultipleAccess with Collision Detection (CSMA/CD) principle to arbitrate access to the bus.Transmission speed is normally 100 Mbit/s. This bus can transfer data using manydifferent protocols (even at the same time), e.g., TCP/IP.

In designing a new and modern control equipment, it’s necessary to use field busses. Thetwo field bus types that are used are CAN and TDM.

The CAN bus is used for communication in both directions between the main computerand digital I/O circuit boards. The transfer speed of a CAN bus is less than that of a TDMbus.

3-16

The TDM bus is single direction and used for high-speed measurement signals. The TDMbus operation status is continuously monitored by the receiving nodes in the control andprotection system, and detected faults will give alarm.

3.4.3 Internal supervisionPeriodic maintenance is minimized by the extensive use of self supervision built into allmicroprocessor-based electronic units, and by the possibility to check all measured valuesduring operation without disturbing the operation.

The internal supervision of microprocessor-based systems includes auxiliary powersupervision, program execution supervision (stall alarm), memory test (both program anddata memory) and supervision of the I/O system communication over the field busses.The operation of the field busses is monitored by a supervisory function in the control andprotection system that continuously writes and reads to/from each individual node of thesystem.

Another example of integrated self-supervision is the switch control unit. In this unit theoutputs to the breakers are continuously monitored to detect failure of the output circuitsof the board.

3.4.4 Automatic voltage controlThe automatic voltage control consists of a closed-loop voltage regulator formed by apositive sequence voltage response, a PI-regulator with variable gain, and the control rulegenerator. The voltage reference signal from the HMI is transformed into a reference forthe voltage regulator. The reference range is limited by parameters and indicated on theHMI for operator feedback. Feedback for the voltage control is the primary voltage,which is measured from the high voltage bus. The regulator output is a susceptancereference value further distributed as an input to the control pulse generator.

3.4.5 Gain supervisorThe control system provides a gain supervisor function for supervision of the SVC MVAroutput. Upon large changes of the impedance in the connecting network the SVC reactivepower output may start to oscillate. This can be explained by high preset regulator gainversus new power system impedance. For oscillations detected in the susceptancereference, Bref , the gain supervisor will automatically reduce the voltage regulator gainuntil the SVC output becomes stable again. When this occurs an alarm will be given andthe gain can manually be reset to normal value from the HMI.

3.4.6 System voltage measurementThe main objective of the data acquisition unit, DAU, is to measure the voltage responseon the primary side of the main transformer. The voltage response, which is fed to thevoltage regulator, is processed in the DAU in order to meet the dynamic demandsregarding speed and stability.

3-17

If a TCR is operated with symmetrical firing, the true voltage response fed to the closedcontrol loop should not contain negative sequence components or harmonics other thanfundamental. On the other hand, if the task is to control unsymmetries, the TCR musthave different controllers for positive and negative sequence voltage components.Therefore an α/β-transformation is employed in order to transform the three-phasevoltage into a rotating vector system in the alpha/beta plane, a so-called space vectorrepresentation (see Appendix C). The voltage space vector is thereafter fed to a functionthat can extract both the positive and negative sequence components from the voltagespace vector.

3.4.7 Control pulse generatorThe main objective of the Control Pulse Generator, CPG, is to generate control pulses forfurther distribution to the Valve Control Unit, VCU. The most important input quantitiesare the susceptance reference from the voltage regulator and the measured SVC-busvoltage. The susceptance reference serves as the control reference value from the voltageregulator while the SVC bus voltage is used for synchronization of triggering pulses andsimulation of TCR and TSC current.

The other control functions are as follows:

• power oscillation damper

• control of external devices

• loss minimization functions

• TCR direct current control

• sequence control of breakers

• protective control functions

• undervoltage control strategy

• supervision of faults in the thyristor triggering system

3.4.8 Operators interfaceThe operator interfaces are provided by workstations for local and remote control (Figure3-8). These are typically Windows NT computers interfaced to the main computer via thelocal area network, LAN.

3-18

COMPAQDESKPRO

Multiscan 20 seTrinitron

Fig. 3-8. Operator work station.

The main functions are:

• Full graphic status displays of various views.

• Display and adjustment of protection settings and control parameters.

• Alarms—all events classified as alarms in order of severity.

• Fault list—all persistent alarms in chronological order.

• Sequence of events—all events/alarms including logging of orders.

The operator interface may also provide high performance transient fault recording.

3.4.9 Remote interrogationRemote interrogation of the control system may be provided by modem communication.An on-line graphical debugger allows the user to view several graphical programmingtool drawings at the same time and inspect any internal software “signal” in real time byjust double-clicking on the line that represents the signal. This fact makes the graphicaldebugger a very useful not only for monitoring, but also for maintenance and debugging.The graphical debugger also allows all thresholds, setpoints, and timer settings to beeasily displayed in various formats (e.g., as tables).

3.4.10 Valve controlThe Valve Control Unit (VCU) is the electrical/optical interface between the firingcontrol system and the thyristor valves. The VCU is realized by two special boards givinga compact design.

3-19

3.4.11 Application software developmentThe application software for the MACH 2 control and protection system are producedusing a fully graphical code generating tool called HiDraw. It is Windows-based softwarethat is very easy to use as it is based on the easiest possible select, drag, place method. It’sdesigned to produce code either in a high level language (PL/M or ANSI standard C) or inassembly language. For functions not available in a comprehensive library (one for eachtype of processor board) it’s very easy to design a new block and link to the schematicwith a simple name reference.

A schematic drawn in HiDraw consists of a number of pages. One page specifies cycletimes and execution order of the other pages. HiDraw includes an on-screen reasonabilitycheck of the drawn schematic, and automatic cross reference between the pages. Asoutput, it produces code and a “make” file ready to be processed.

The next step in the workflow is to run the make file (on the same computer) whichmeans invocation of the necessary compiler/assembler and link locate programs (usuallyobtained from the chip manufacturers). The result is a file that is ready to be downloadedfrom the computer to the target, and stored in the flash PROMs.

3.4.12 Debugging facilitiesFor debugging, a fully graphical debugger operating under Windows is used. Thedebugger allows the operator to view several HiDraw pages at the same time, and look atany internal software “signal” in real time by just double clicking on the line thatrepresents the signal. Parameters can easily be changed by double clicking on their value.As a complement, a fully symbolic debugger is available either on a computer or a dumbterminal.

For fault tracking, it’s easy to follow a signal through several pages because when asignal passes from one page to another, a double click on the page reference willautomatically open the new page and allow the trace to continue immediately on the newpage. There are also a number of supporting functions such as single or multiple steppingof tasks (one page is normally a task) and coordinated sampling of signals.

The fact that the debugger allows inspection of signals while the application is runningmakes it very useful not only for debugging but also facilitates maintenance. Because thedebugger works in the Windows environment, it’s also possible to transfer sets of signalvalues to other Windows-compatible programs, such as Excel, for further analysis.

3.5 Trends in Digital ControlElectronics have evolved at an astounding rate in the last years. It’s very difficult to makelong-term predictions, although trends are apparent. Basically it can be said that:

• Availability of resources from digital controllers will shorten the time between thedevelopment of a new control law and its practical implementation.

• The improvement in speed and reliability of the communication channels will allowthe creation of safe methods for remote commissioning and maintenance.

3-20

• The adoption of common use, more flexible hardware and software is an observedtrend, making systems integration much easier and lowering maintenance costsbecause of high scale production.

• An easy access to a superior hierarchical level can be provided by object-orientedtechnologies. For instance, the use of OPC (Object Linking and Embedding forProcess Control), where the controller opens a window in a higher level supervisorysystem, and makes available all of its resources (adjustments and commands) withoutthe system integrator worrying about the knowledge of the controller implementationdetails.

• On-site implementing upgrades should be easier, because no hardware changes willbe needed. On the other hand, design and documentation efforts demanded bysoftware modification could be large.

• PC-based systems are becoming more cost-effective, and have been occupyingtraditional PLC space.

• The costs will drop, as a result of the electronic circuit large scale integrationincrease, as well as new technologies and the enhancement of software techniques.

• There will be a need to develop more system analysis tools to handle the largediversity of control laws performing in different machines of the system.

References3-1 Karl J. Åström and Björn Wittenmark, Computer-Controlled Systems, Theory and

Design, Second Edition, 1999.

3-2 G.S. Virk, Digital Computer Control Systems, Macmillan New Electronics, 1991International Editions, 1999.

3-3 G. Olsson and G. Piani, Computer Systems for Automation and Control, PrenticeHall, 1992.

3-4 IEEE, Guide for the Preparation of Excitation System Specifications, Std 421.4-1990.

3-5 Corbetta and G. Ottaviani, “Digital Measurement Procedures in a Microprocessor-Based Excitation System,” EPE, Florence, 1991.

3-6 W. S. Levine, The Control Handbook, CRC Press, 1995.

3-7 Phillip A. Laplante, Real-Time Systems Design and Analysis, IEEE Press.

3-8 S. Corsi, M. Pozzi, and G. Tagliabue, “A New Digital Simulator of the Turbine-Alternator-Grid System (STAR) for Control Apparatus Closed-Loop Tests,”IEEE/PES Summer Meeting, Berlin, 1997.

3-9 V. Arcidiancono, S. Corsi, G. Ottaviani, S. Togno, G. Baroffio, C. Raffaelli, andE. Rosa, “The ENEL’s Experience on the Evolution of Excitation Control

3-21

Systems through Microprocessor Technology,” IEEE Transactions on EnergyConversion, Vol. 13, No. 3, pp. 292–299, September 1998.

3-10 ABB Power Systems, MACH 2 Description, Technical Report RU 8037 AU,1999.

Chapter 4

State-of-the-Art in Intelligent Controls

Deregulation requires that utilities exercise less conservative operation regimes and moreprecise power-flow control. This is possible only by monitoring and controlling thesystem in much more detail than is, or has been, the case in present and past practice.

The large quantity of information required can be provided in many cases throughadvances in telecommunications and computing techniques. There is still the need forevaluation techniques that extract the salient information from the large amount of rawdata to use for higher-order processing. Up until now, the extraction of qualitativeinformation is still done by the human expert, who can be overwhelmed in emergencysituations when fast decisions are needed. The future operators also need to have theability to specify the operating strategy in qualitative form, which is then translated intoquantitative form in order to be processed by the computer control.

One of the main motivations for using intelligent systems is to provide this importantinterface between qualitative and quantitative information. Beside the control-centerapplications, intelligent control can be applied in a decentralized manner. For exampleconsider closed-loop generator control. A consideration with existing control methods isthat the control law is based mainly on a linearized model and the control parameters aretuned for certain operating conditions.∗ In case of a large disturbance, the systemconditions will deviate significantly from the linearized condition, and the controllerparameters may no longer be valid. In this case the controller may even add adestabilizing effect, such as negative damping.

Intelligent Systems can be categorized as:

• Expert Systems (ES) which process qualitative as well as quantitative knowledge withemphasis on the qualitative results.

• Fuzzy Systems (FS) which quantify qualitative knowledge including uncertainties.

• Artificial Neural Networks (ANN) which infer quantitative information throughapproximation techniques and classify quantitative data into higher-order qualitativecategories.

• Decision Trees (DT) which classifies quantitative data into discrete sets of qualitativecategories.

Expert System techniques are often associated with the software engineering concept ofintelligent computing environments. Data and rules are formulated on a symbolic level inpseudo-natural language. In the ideal case, the “reasoning process,” i.e., the formulationof goals and the subsequent application of rules, are transparent to the user. Heuristic

∗ Control is typically verified by nonlinear simulation for a limited number of operating conditions and

disturbances.

4-2

reasoning (inspired by rules of thumb) are implemented in order to limit the number ofbranches of the decision tree to be exploited during the reasoning (i.e., deduction)process. Due to the nature of this approach, expert system techniques are often discussedin the context of an intelligent user-friendly human-machine interface, where not only realdata and network topology maps but also abstract reasoning concepts like rules anddecision trees are displayed graphically [4-1].

Expert system techniques are therefore usually implemented as off-line decision aids.Reference 4-2 discusses a voltage-control expert system for the off-line changes of on-load tap changer settings. It specifically draws attention to the fact that the heuristicnature of the off-line control rules limits their range of validity. Other examples ofapplications of expert systems for power system off-line monitoring and control can befound in reports published by several task forces of CIGRÉ WG 38.06 [4-3–7].

In the following we will concentrate on the applications of Fuzzy Systems, ArtificialNeural Networks and Decision Trees to power system control.

4.1 Fuzzy Systems for Power System ControlFuzzy sets and systems were first introduced by Zadeh [4-8]. Fuzzy systems come in twoflavors:

• Empirical or rule-based fuzzy systems

• Self-adaptive fuzzy systems (self-organized or unsupervised fuzzy systems)

In the literature, fuzzy sets and fuzzy control are mostly discussed in terms of qualitativeattributes like cold or warm and qualitative rules like “if temperature is cold with alikelihood of 0.7 then increase heating fast.” These empirical rules are often establishedfrom existing expertise in manual control and the corresponding fuzzy systems arereferred to as empirical fuzzy systems.

However, in the area of power system control, as for example power system stabilizers,this expertise may not exist for unusual operating conditions. It’s therefore necessary toestablish the fuzzy sets and rules in a more systematic, autonomous manner and thecorresponding fuzzy systems are referred to as self-adaptive fuzzy systems.

Let us briefly illustrate these concepts by looking at the example of fuzzy temperaturesets [4-9]. If the initial input set is the range of temperatures from 0oF to 120oF, themembership function describing the three fuzzy sets cold, warm and hot may be centeredat a1 = 40oF, a2 = 70oF and a3 = 100oF, and have a triangular shape and a maximal width σ of 20oF as shown in Figure 4-1.

Instead of defining center, shape and width of the membership function by empiricalrules, one can choose a more systematic approach using data analysis. For example, in thecase of load forecasting, sampling of the load data might indicate that the load exhibitsthree different behaviors correlated with the temperature. A clustering algorithm mighthave identified three typical temperatures a1, a2, and a3, with the width of the clusterdefining the width of the membership functions µ1, µ2 and µ3.

4-3

4020 60 10080 120

1

µi(e) i = 1, 2, 3

Temperature e[F]a1 a2 a3

30 50 70 90 110

Fig. 4-1. Membership function for fuzzy temperature sets.

In addition, one can choose a Gaussian function, which is continuously differentiable,instead of the triangular (or the sometimes used trapezoidal shape) without altering thedegree of membership of any given temperature significantly.

Whether one defines the membership function empirically or self-adaptively, there arealways some degrees of freedom; for example, the number of fuzzy sets and membershipfunctions.

As an analogy to crisp sets, one can define union, intersection and complement of twofuzzy sets A and B by defining the membership functions corresponding to union,intersection and complement. One can further define fuzzy rules either by establishingthese rules empirically or in a self-adaptive manner.

Finally, a mapping from a crisp number to the fuzzy set can be defined consisting of thisnumber only (singleton fuzzification). Also a mapping from a fuzzy set onto a numbercan be defined by choosing this number as the center of average of the integral defined bythe fuzzy membership function (center of average defuzzification). Figure 4-2 shows thestructure of fuzzy system.

For the purpose of power system control it is sufficient to note that the fuzzy system is amapping

F: Un ⊆ ℜ n ->ℜ , F(e) = u

This mapping F will be constructed as an approximation to the controller φ(e,t).

It is shown [4-9] that there is a class of self-adaptive fuzzy systems F with Gaussianmembership functions φj that can be written in a closed form as:

Self-adaptive fuzzy systems given in closed form have the advantage that stabilityanalysis can be performed and tasks like optimal control can be addressed.

Self-organizing fuzzy controllers therefore fall into the class of adaptive controllers andthe related stability issues can be explored with adaptive control techniques. Stability ofpower system controllers is discussed in more detail in reference 4-10.

∑=

==m

1jjj )(b)F(u ee φ

4-4

Fuzzy Rule Base and Fuzzy Inference EngineIF e is A AND e1 is A1 ... AND en is An THEN u is B

Fuzzifiere → Fuzzy Set (A, µA)

DefuzzifierFuzzy Set (B, µB) → u

e ∈ ℜ n

A

B

u ∈ ℜ

Fig. 4-2. Structure of a fuzzy system.

Neither approach to fuzzy systems necessarily needs a detailed state-space model of thecontroller. The advantage of the empirical approach is that heuristics and humanknowledge can be incorporated. However, the demonstration of stability for this type ofcontroller is very tedious if not impossible.

4.1.1 State-of-the-art of fuzzy control for power systemsWe now give an overview of studies of fuzzy systems in the area of power system orgeneration control [4-11].

The majority of fuzzy controllers can be found in the area of excitation control, especiallypower system stabilizers (PSS). An upcoming important area is control of powerelectronic devices. Although the majority of investigations perform feasibility studiesusing computer simulation only, several authors study the implementation of the fuzzycontroller on a PC or DSP in order to control actual small generators or motors in alaboratory environment. In most cases, the membership functions are established basedon data samples.

The comparison of fuzzy controllers and conventional controllers stresses advantages offuzzy controllers as being “generic” parametric models instead of circuit-based statespace models. The self-adaptive controllers can be easily tuned to different operatingconditions, and all projects report better tracking capabilities of the fuzzy controllerscompared to conventional controllers.

However, the sensitivity issues concerning the range of validity of the tuning and thedetection of changes of operating conditions still needs to be investigated forconventional as well as for fuzzy controllers. This is especially important for powersystem control where topology, load, and generation can change stochastically anddiscontinuously.

A lot of progress has been made concerning the application of fuzzy systems to powersystem control problems. For feasibility studies, most authors experiment with empirical

4-5

rules and data. A few projects, using self-organizing techniques, however, have beeninstalled on a microprocessor and tested in a research lab environment either in academiaor a utility. The next section describes an operational application of fuzzy control in a realpower system.

Hassan, Malik and Hope applied the fuzzy logic control (FLC) to PSS design [4-12]. Inthis method, the output stabilizing signal was calculated based on the representation ofthe alternator state in the phase plane. Hiyama, Kugimiya, and Satoh proposed PID typefuzzy logic PSS [4-13]. They took into account the PID information of the generatorspeed. Additional parameters were also tuned off-line to minimize the performance index.Recently, the self-organizing Fuzzy Auto-Regressive Moving Average (FARMA)controller was studied to enhance the low frequency damping of a synchronous machine[4-15]. In contrast with a conventional FLC, where the rule base and membershipfunctions are supplied by an expert or tuned off-line through experiment, the FARMAFLC needs no expert in making control rules. Instead, rules are generated using thehistory of input-output pairs. The generated rules are stored in the fuzzy rule space andupdated on-line by a self-organizing procedure.

4.1.2 Implementation of fuzzy logic PSSIn joint research, Kumamoto University and the Kyushu Electric Power Companyproposed a microcomputer-based fuzzy logic power system stabilizer (FLPSS) to enhancepower system stability through control of thyristor exciters. Through simulation studies,experiments on a 5 kVA laboratory system, and implementation on an actual 5 MVAhydro unit, the effectiveness of the FLPSS was demonstrated [4-13]. In addition, a two-year evaluation of the FLPSS was finished in March 1996 on 30.2 and 23.4 MVA hydrounits in the Kyushu Electric Power System [4-14]. Damping of oscillations weresignificantly increased. The FLPSS has been in service since June 19, 1997 on a hydrounit with the rating of 90 MVA at the Hitotsuse Hydro Power Station in the KyushuElectric Power System.

The proposed fuzzy logic power system stabilizer (FLPSS) is set up by using amicrocomputer with AD and DA conversion interfaces. All the signal conditioning andthe generation of stabilizing signals are performed by the on-line microcomputer. SeeFigure 4-3.

4.1.3 The future of fuzzy logic power system stability controlsThere is continued debate on the fuzzy versus conventional control (reference 4-59 isentertaining and instructive). Although the fuzzy logic power system stabilizers are fieldtested as described above, there is limited experience, even in the simulation world, offuzzy logic power system stability controls in large power systems with multiple,interacting oscillation modes. Experience with the more sophisticated types of fuzzy logiccontrol is even more limited.

Although most of the literature on power system fuzzy logic control is on replacement ofconventional control, many actual industrial applications (in other industries) are forhigher level or supervisory control [4-60,4-61]. In power systems, fuzzy logic controls

4-6

may be attractive for higher level, nonlinear, and discrete controls, rather than asreplacement of essentially linear continuous controls.

4.2 ANN for Power System ControlArtificial neural networks have been applied in technical areas since the early 1960s,when Widrow and Hoff developed an adaptive least square estimator called ADALINE.ANNs come in two major categories:

• supervised ANNS,

• unsupervised ANNs.

Supervised neural networks perform approximation tasks using a special combination ofnon-linear basis functions called sigmoid functions. They therefore solve problemssimilar to problems solved by regression and parameter estimation techniques.

VoltageDetectors

Timer

AVR Exciter

UPS

D/A Micro-

computer

A/D Power

Transducer

PQVF

WT

G

Protection Unit

Monitoring Unit

AC100V

AC100V

FLPSS

‚ r

Fig. 4-3. Basic configuration of PSS prototype and its overview.

In this framework, classification tasks can be formulated as the task of finding aregression model for the function which maps an input vector x onto its class label, forexample TRUE or FALSE coded with binary numbers.

The multi-layer perceptron (MLP) is probably the most heavily investigated supervisedANN model. It can be used in nearly every area of power systems where a task can beformulated as an approximation problem. As a classifier (approximation of the Booleanfunction secure/insecure or trip/no-trip signal) it is applied in power system securityassessment [4-16] and on-line security control to initiate load shedding at a bus [4-17].

The MLP is often used in combination with Fuzzy Systems where qualitative attributeslike hot or cold temperatures are first translated into numbers. The MLP is then used as aregression tool in order to estimate additional parameters [4-18].

4-7

Unsupervised networks reduce the complexity of the data sets by either reducing thedimensionality of the input data or by grouping input data into categories of “typical” dataand by constructing a typical presentation (code vector) for each class. Unsupervisedneural nets fall into the same class of tools as statistical non-parametric data analysis,clustering algorithms, and encoding or decoding techniques.

Unsupervised ANNs which quantize data into categories provide a choice of freeparameters. The ART networks fixes the radius of the class but allows a variable numberof classes, whereas Kohonen’s self-organizing feature map fixes the number of categoriesbut allows varying class sizes.

In the area of power system security assessment the ART network [4-19] and theKohonen map [4-20] are used to reduce the space of all feasible operating points into afinite set of typical operating points.

Unsupervised ANNs are often used in combination with supervised approaches orconventional tools. The unsupervised net serves as pre-processing tool for data reductionand the supervised net estimates associated parameters like security classes [4-21,4-22].

4.2.1 ANN applicationsIn the 1970s simple ANN-based machine-learning techniques were explored for transientstability [4-23]. With the emergence of more powerful computers, ANN gained renewedinterest from 1988 on, when Sobajic et al. [4-24], and Aggoune et al., [4-25] assessedtheir potential for transient stability and static security assessment. These projects haveled to a sudden upsurge in applying neural net approaches to many power systemproblems. A bibliographical survey covering 1988–1993 world-wide is presented in thepaper by the CIGRÉ Task Force 38.06.06 on Artificial Neural Net Applications in PowerSystems [4-4]. This survey was updated by Niebur and Dillon [4-26] based on a review ofmore than 400 publications regrouped into 200 different projects published before April1995.

Time-series prediction in the area of load forecasting has been one of the most examinedareas for ANN applications. It was mainly motivated by the lack of automated tools in theutilities and by the expected economic gain. Research in other major application areaslike security assessment attempts to exploit the data reduction, classification, andregression capabilities of ANN in combination with conventional simulation techniques.The potential of ANNs for non-linear adaptive filtering and control stimulated research inthe area of control of highly non-linear power system behavior.

For power system control, the control tool, whether conventional or ANN has to beoperated on-line. Available reaction time is extremely limited and control errors caneasily lead to a breakdown in a substantial portion of the interconnected system.Therefore power system control is still done in the most conservative manner. In criticalsituations, it’s the practice of some experienced operators to even remove conventionalcontrollers like power system stabilizers. New control tools need to be extensively testedbefore they can be integrated into the existing complex power system. Field tests for

4-8

control, however, have been reported for isolated components like photovoltaic storage[4-27].

Similar remarks apply to the area of security assessment. Further, in both areas, datacovering significant periods of operation are not readily available and have to be collectedfor the specific ANN applications.

In the area of control, field tests are reported by Kumamoto University and SanyoElectric, Japan [4-27]. For fast dynamic security monitoring in a medium scale networkwith diesel and wind power production, a pilot installation is running successfully in theisland of Lemnos, Greece [4-28].

4.2.2 ANN application in security assessmentSecurity assessment can be divided into two levels: classification and boundarydetermination. Classification involves determining whether the system is secure orinsecure under pre-specified contingencies. Classification does not in itself indicatedistance from the operating condition to the insecure conditions. Boundary determination,on the other hand, involves quantifying this distance. A boundary is represented byconstraints imposed on parameters characterizing pre-contingency conditions. These pre-contingency parameters are called critical parameters. Once the boundary is identified,security assessment for any operating point can be given as the “distance” between thecurrent operating point and the boundary. Assessment in terms of pre-contingencyoperating parameters instead of the post-contingency performance measure is moremeaningful to the operator as it directly identifies the parameters to control, as well ashow to adjust them, in order to maneuver the system with respect to security boundaries.

In many North American utilities, the traditional boundary characterization is a two-dimensional graph called a nomogram [4-29–31]. To develop a nomogram, two criticalparameters are chosen and all other critical parameters are set to selected values within atypical operating range. The non-critical parameters are set to constant values. Points onthe nomogram curve are determined by repeating computer simulations, varying onecritical parameter while keeping the other constant. The main disadvantages of thisapproach include intensive labor requirement, inaccurate boundary representation, andlittle flexibility in integrating with the energy management system (EMS). The inaccuracyof the nomogram results mainly from linear interpolation between boundary points andinsufficient information contained in critical parameters. An ANN technique has beenused in a security boundary visualization method to overcome these disadvantages [4-32,4-33].

The procedure for boundary visualization consists of the following major steps:

1. Security problem identification: Identify the specific set of security problems to becharacterized and operating parameter candidates that may have influence on them.

2. Base case construction: Construct a base case power flow solution that appropriatelymodels the system conditions.

4-9

3. Data generation: Automatically generate a database with each record consisting ofpre-contingency operating parameters and the corresponding post-contingencymeasure.

4. Feature selection: Select the best subset of pre-contingency operating parameters foruse in predicting the post-contingency performance measure.

5. Neural network training: Train a neural network using the selected parameters and thedatabase to map the relationship from the pre-contingency operating parameters to thepost-contingency performance measure.

6. Visualization: Provide an easily understood automatic visualization of the securityboundary in the space of operating parameters that can be monitored and controlledby the system operator.

Data generation is a very important step [4-34]. The ultimate boundary captured by thewhole procedure will characterize the data that is provided to the neural network. If thisdata does not reflect what actually occurs in system operations, the boundary will beincorrect. A systematic method, call ASAS [4-35] has been developed to generate thedata for neural network training. This data consists of a large number of samples, witheach sample corresponding to a simulation of the same contingency but for differentoperating conditions, and consisting of values for pre-contingency operating parameterstogether with the post-contingency performance measure. This data is used to train aneural network to compute the post-contingency performance measure R as output giventhe pre-contingency operating parameters x as input, resulting in the relation R = f(x),where f represents the neural network mapping function. Standard MLP networks havebeen used for this application.

Once the neural network is trained, the relationship between the post-contingencyperformance and the pre-contingency operating parameters can be inverted, subject to thepower flow equations, in identifying the boundary. That is, the problem of boundaryidentification is solved by finding x that simultaneously satisfies:

f (x) - Rb = 0 (1)

h ( u) = 0 (2)

where (1) represents the neural network mapping function, (2) represents the power flowequations, x is the critical parameter vector, Rb is the threshold value of R, and u is theinput parameter vector to the power flow program. The vector x may include bothindependent critical parameters (e.g., real power injections) and dependent criticalparameters (e.g., flows), and is therefore a function of u. Because the presentedparameters (those corresponding to the two coordinate axes) must be varied in drawingthe boundary, the influence of these variations on dependent critical parameters should beconsidered accordingly.

For visualization of an individual boundary, i.e., the boundary for a single securityproblem under a given contingency, the computation used in solving equations (1) and (2)is based on a derived form of the neural network mapping function, expressed as

4-10

f (z, gy (y0, ∆z1, ∆z2)) - Rb = 0 (3)

where x=[z,y], z is the independent critical parameter vector, y is the dependent criticalparameter vector, y0 is the dependent critical parameter vector corresponding to a specificoperating condition, and gy models the influence of the z1 and z2 changes on thedependent critical parameters y, where z1 and z2 represent the two presented parameters.The visualization algorithm starts from the minimum value of z1 and solves equation (3)for z2. Then it increases z1 by a fixed step, updates y, solves for z2, and repeats until itreaches the maximum of z2.

In visualizing a boundary comprised of two or more constraints, we proceed as follows.As shown in Fig. 4-4, for each interval ∆z1, we first identify the two individual boundaryfunctions that are binding for the composite boundary. To do this, we rank the functionsin descending order of z2. For each pair of neighboring functions in this rank, we checkan arbitrarily selected point (marked with crosses) between them to see if it is secure forall security constraints. If so, this point is inside the secure region, and the correspondingneighboring individual boundary functions must be the binding functions for thecomposite boundary for this interval. The composite boundary is therefore identified asthis pair of individual boundaries. In the next interval, if there are no other individualboundary functions between the two binding functions identified in the previous interval,then these functions are also binding for the new interval. In this case, it is not necessaryto perform the check for this interval. Once it is no longer possible to find any securepoint, then the algorithm stops.

z1

B2

B1

z2B3

Fig. 4-4. Algorithm illustration for composite boundary visualization.

4.2.3 ANN application in power system stabilizationControl of large-scale systems such as power systems has been recognized as a foremostchallenges in control engineering due to its nonlinearity and complexity. The use of anartificial neural network is very attractive because of its nonlinear mapping ability. Forcomplexity coming from high dimension or from the spatial distribution of a large-scalesystem, decentralized control is a practical approach. Neural networks have attractivecapacity in handling sensory information, and performing collective learning from thedata sets given for a subsystem in the decentralized control approach. The approximationproperty of neural networks can make it possible to organize subsystem dynamics to a

4-11

certain degree by training the input/output relationships obtained in the full systemoperation. From this point of view, a neural network based power system stabilizer can bedesigned for a large-scale power system when only local input/output information data fora subsystem, i.e., power plant data, is available.

A practical power system stabilizer to enhance the damping of the low-frequencyoscillations must be robust over a wide range of operating conditions. However,conventional PSS design approaches based on linearization around the normal operatingpoint have deficiencies and difficulties coming from nonlinearities in the system.Recently, neural networks have been investigated for power system stabilizing control.Most cases are limited to speed deviation control with supplementary excitation signal fora single generator–infinite bus system.

Difficulties in a power system stabilizer design come from the handling of nonlinearitiesand interactions among generators. During the low-frequency oscillation, rotor oscillatesdue to the unbalance between mechanical and electrical powers. Electrical power hasnonlinear properties, and this is a key variable affecting the rotor dynamics. Thus,handling the nonlinear power flow properly is the key to the PSS design for a multi-machine power system. The use of neural networks’ learning ability avoids complexmathematical analysis in solving control problems when plant dynamics are complex andhighly nonlinear.

Neural networks in control has mainly used Model Reference Adaptive Control (MRAC)[4-36–40]. However, the MRAC approach has difficulty in selecting an appropriatereference model. Recently, a general purpose controller, an Optimal Tracking Neuro-Controller, was developed to minimize a general quadratic cost function of tracking errorsand control efforts [4-41]. This results in a hybrid of feedback and feedforward neuro-controllers in parallel. The feedforward neuro-controller (FFNC) generates the steady-state control input to keep the plant output to a given reference value, and the feedbackneuro-controller (FBNC) generates the transient control input to stabilize error dynamicsalong the optimal path while minimizing the cost function. A novel inverse mappingconcept is developed to design the FFNC using a neuro-identifier. The use of generalquadratic cost function provides “optimal” performance with respect to trade-off betweenthe tracking error and control effort. Since the cost function is defined over a finite timeinterval, a Generalized Backpropagation-Through-Time (GBTT) algorithm wasdeveloped to train the feedback controller.

Optimal tracking neuro-controller. We consider a system in the form of the generalnonlinear auto-regressive moving average (NARMA) model:

y f y y y u u uk k k k n k k k m( ) ( ) ( ) ( ) ( ) ( ) ( )( , , , , , , , )+ − − + − − += ⋅ ⋅ ⋅ ⋅ ⋅ ⋅1 1 1 1 1 , (4)

where y and u , respectively, represent output and input variables, k represents timeindex, and n and m represent the respective output and input delay orders.

The above control objectives can be achieved by minimizing the following well-knownquadratic cost function:

4-12

J = − + −∑ +=

1

21 2 2

1( ( ) ( ) )( ) ( )Q y y R u uref k ref k

k

N, (5)

where yref is a reference output, uref is the steady-state input corresponding to yref , andQ and R are positive weighting factors. This quadratic cost function or performanceindex not only forces the plant output to follow the reference, but also forces the plantinput to be close to the steady-state value in maintaining the plant output to its referencevalue.

An optimal tracking neuro-controller (OTNC) is designed with two neuro-controllers inorder to control a nonlinear plant that has a non-zero set point in steady-state [4-41]. Afeedforward neuro-controller (FFNC) is constructed to generate feedforward control inputcorresponding to the set point, and trained by the well-known error Backpropagationalgorithm. A feedback neuro-controller (FBNC) is constructed to generate feedbackcontrol input, and trained by a Generalized BTT (GBTT) algorithm to minimize thequadratic performance index. An independent neural network named neuro-identifier isused when the above two neuro-controllers are in training mode. This network is trainedto emulate a plant dynamics and to backpropagate an equivalent error or generalizeddelta [4-36] to the controllers under training. Fig. 4-5 shows an architecture for theoptimal tracking neuro-controller for a nonlinear plant. In the figure, the tapped delayoperator ∆ is defined as a delay mapping from a sequence of scalar input, }{ )(ix to a

vector output with an appropriate dimension defined as )...,, ,( )()2()1()1( pxiii xxxx −−−− =�,

where p = n for the output variable y, and p= m-1 for the input variable u.

u (k-1)

(k-1)∆

∆

Σ

yref

y ref

y(k)

y(k)

fbu u

uff

(k) u (k)

Neuro-ControllerFeedback

Neuro-ControllerFeedforward

Plant Dynamics

Neuro-Identifier

y(k+1)

y(k+1)

y(k)

u

(k)

Tapped Delay Operator

u (k-1)

u (k)

Fig. 4-5 Block diagram for the optimal tracking neuro-controller.

The study power system. The neuro-controller is applied to a 5-bus power system [4-42]to stabilize low-frequency oscillations (Figure 4-6). The power system consists of threepower plants: two are thermal units and one is a hydro unit. The power system hassustained low-frequency oscillations due to disturbances. The control objective is toimprove system damping by using a supplementary excitation control applied to thesecond generator.

4-13

G2

( Hydro Plant )

2

L3

3 4

50.2+j0.1

1.0+j0.15 0.6+j0.15

0.8+j0.2

0.02+j0.06(0.03)

0.08+j0.24(0.025)

0.06+j0.18 (0.02)

0.06+j0.18(0.02)

0.01+j0.03(0.01)

0.04+j0.12(0.015)

0.4+j1.2(0.02)

G1

( Thermal Plant )

1Voltage : 1.06+j0.0

L4

L2 L5

Install PSS

Power Flow

Power: 0.75+j0.5

G3

( Thermal Plant )

0.67+j0.67

Fig.4-6. The power system with 3 generators and 5 buses.

Typical IEEE governor and turbine models are used: TGOV1 (2nd order) for the thermalplant and IEEEG2 (3rd order) for the hydro unit [4-43]. The IEEE exciter and voltageregulator model EXST1 (4th order) is used for this study on which supplementaryexcitation control input is to be injected. As a result, a 9th order model for thermal plantsand a 10th model for the hydro plant are used to represent the nonlinear characteristicsand the low-frequency oscillations in simulations.

Training of the neural networks. The Optimal Tracking Neuro-Controller is applied toGenerator 2 to provide supplementary excitation signal as a power system stabilizer.Since the output variables, frequency, angle, and the power flow, are all deviations fromthe respective references, the feedforward controller was not used. The training patternsof the Neuro-Identifier are generated by the power system simulations starting from thesteady-state initial value in a wide range of operating conditions and randomly generatedcontrol inputs history within the conventional PSS operation region. During the low-frequency oscillation in the range of 1~2 Hz, it’s assumed that the exciter can beapproximated as a second-order model. Therefore, the Neuro-Identifier is constructed toemulate the power flow dynamics as a third-order model that includes the dynamics ofexciter and the excitation field voltage. The discrete-time training patterns are obtainedwith the time step of 0.04 sec in simulation. This allows at least twenty sampling pointsin a cycle of the low-frequency oscillation under 1.25 Hz.

The Neuro-Identifier consists of one hidden layer with 40 nodes, an input layer with 7input nodes and an output layer with one node. Three of the seven input nodes are for itsoutput history,

)2()1()( , , −− ∆∆∆ kkk PePePe ; two are for control input history, )1()( , −kk uu ;

and two for )()( , kk δω ∆∆ . The Neuro-Controller has one hidden layer with 40 nodes, an

input layer with 6 input nodes and an output layer with one node. Three of the six input

4-14

nodes are for output history, )2()1()( , , −− ∆∆∆ kkk PePePe ; one is for previous control input

)1( −ku and two are (k))( , δω ∆∆ k . The cost function for the N-step ahead controller is set

with the weightings Q = 1.0 and R = 0.02.

To avoid oscillation during training stage, weight parameters in the Neuro-Identifier arecorrected with the average of corrections calculated for ten patterns. Training of theNeuro-Controller is done in two phases. First, training is done with a small N ( = 3) sincein the beginning it has little knowledge of control. A small number of steps prevents thesystem from diverging. Training is carried on with a gradually increasing N until itreaches 8 so that the system can be controlled for a longer duration of time. Then, trainingis carried on with N fixed at 8. It takes about 30 minutes on an IBM-PC 486 computer totrain two neural networks: the Neuro-Identifier and the Neuro-Controller.

Comparison of the control results. Figure 4-7 shows the speed deviation of Generator 2for a three-phase ground fault at midpoint of a half the line 4–5, which cleared after 0.2sec. The figure compares the cases without a control and with supplementary excitationcontrols by the conventional PSS, STAB4 [4-43], and the Neuro-PSS.

Time [Sec]

Without Control STAB4 Neuro-PSS

Speed-dev. of the 2-nd Gen. ( 0.75[p.u.] )

[Hz]

-0.5

-0.3

-0.1

0.1

0.3

0.5

0 1 2 3 4 5 6

Fig. 4-7. The speed deviation of generator 2 for the line fault disturbance in a normal loadcondition.

Figure 4-8 shows the speed deviation for the same disturbance when the power system isin a light loading condition (0.5 p.u. generating power) and Figure 4-9 shows speeddeviation for a heavy loading condition (1.0 p.u.). The figures show that both controllerswork very well judging from small swings with large damping. The performance of thecontrollers are compared in Table 1with the integral-time-error (ITE) computed with thecost function (5). Observations from the table show that the Neuro-PSS works very welljudging from the ITE performance in both the heavy or the light load compared to thenormal load condition. The ITE performance of the conventional PSS shows largervariation to loading conditions because the parameters in the STAB4 were optimized inthe normal loading condition.

4-15

Time [Sec]

Speed-dev. of the 2-nd Gen. ( 0.5 [p.u.] )

[Hz]

Without Control STAB4 Neuro-PSS

-0.4

-0.2

0.0

0.2

0.4

0 1 2 3 4 5 6

Fig. 4-8. The speed deviation of generator 2 for the line fault disturbance in a light loadcondition.

Time [Sec]

Without Control STAB4 Neuro-PSS

Speed-dev. of the 2-nd Gen. ( 1.0[p.u.] )

[Hz]

-0.5

-0.3

-0.1

0.1

0.3

0.5

0 1 2 3 4 5 6

Fig. 4-9. The speed deviation of generator 2 for the line fault disturbance in a heavy loadcondition.

Figure 4-10 shows the speed deviation for other disturbances coming from stepwiseloading conditions: 0.15 p.u. increase at 0.24 sec, decrease at 0.96 sec. and cleared at 1.44sec when the power system is in the heavy loading condition. The figure shows that theNeuro-PSS works very well judging from small swings.

4-16

Time [Sec]

Without Control STAB4 Neuro-PSS

Speed-dev. of the 2-nd Gen. ( 1.0[p.u.] )

[Hz]

-0.5

-0.3

-0.1

0.1

0.3

0.5

0 1 2 3 4 5 6

Fig. 4-10. The speed deviation of generator 2 for the load change disturbance in a heavyload condition.

Table 1. ITE performance evaluation for the line fault disturbance

Loading 0.5 p.u. 0.75 p.u. 1.0 p.u.

Without Control 6.04 100(%) 12.03 100(%) 22.24 100(%)

STAB4 1.81 30.0(%) 2.19 8.2(%) 2.83 12.7(%)

Neuro-PSS 1.67 27.6(%) 1.89 15.7(5) 1.92 8.6(%)

4.3 Decision Trees for Power System ControlDecision trees (DTs) are learn-by-example classifiers which are particularly well suitedfor discrete event control [4-44,4-45]. Artificial neural networks (ANNs) can also be usedfor discrete event controls, and they are more general than decision trees. Neural networkscan associate their input vectors with a continuous range of output values, whereasdecision trees are only suited for classification problems having a small number of outputcategories such as stable/unstable. But when a problem can be reduced to a small numberof choices, then decision trees have important advantages. The decision trees reported in[4-46–50] require only a few minutes to train whereas neural networks usually requiremuch more computation for the training. When a particular case is classified by a DT, wecan see which threshold criteria were met, i.e., why the case was classified and how theoutcome would have changed if certain input variables had been different. Anotheradvantage of decision trees is that when you have training data with maybe 250 variablesin each input vector, the DT training algorithm usually selects a much smaller subset,perhaps 25 variables, to be used for classification.

4-17

4.3.1 Relation of angle stability decision trees to on-line dynamic securityassessment

Decision trees have been developed for on-line preventive control and also for real-timeremedial action control. The first research and industrial use of DTs for angle stabilitycontrol was in the area of on-line preventive control [4-51–53]. These DTs are designedto perform on-line dynamic security assessment (DSA). Training sets are extracted fromoff-line simulations of critical contingencies applied to a large number of pre-faultequilibrium conditions. The input vector contains various static parameters from the pre-fault equilibrium point such as key generation and transfer levels. The desired outputreflects whether any of the contingencies caused instability for that equilibrium. The DTsare then used on-line to predict the vulnerability of the power system in its presentequilibrium state to those contingencies.

The DTs for real-time remedial action control [4-46–50] could be trained either from off-line simulations or from on-line simulation tools that are being developed to perform on-line DSA. Power system protection and large-scale stability controls have traditionallyrelied upon off-line simulations that are transformed into decision rules by engineers.Classifier training algorithms can perform the same tasks using large numbers ofsimulations and predictor variables. An emerging possibility is to train the classifiersusing on-line DSA [4-54,4-55]. These on-line simulations can already be used to programdiscrete event controls such as generator tripping (see Chapter 5). The resulting controlsare custom tailored to the current operating conditions. The same simulation capabilitiescould generate the training sets for DTs that perform real-time, remedial action control.

4.3.2 Decision trees for real-time transient stability predictionThe earliest research on DTs for real-time control investigated prediction of angleinstability using synchronized phase angle measurements from all 10 generators in theNew England 39 bus test system [4-46,4-47]. In that work, it was proposed to train DTsoff-line to handle a specific range of operating conditions. Training sets were created bysimulating three-phase faults of various duration on all the buses and transmission lines.Simulated generator angle measurements were taken over an eight cycle windowimmediately after fault clearing. Three successive measurements of the generator angleswere used, and then two velocities and one acceleration were computed from the anglemeasurements of each generator. From this snapshot immediately after fault clearing, thedecision trees correctly predicted whether loss of synchronism would occur in the nextfour seconds with over 97% accuracy. Robustness to variations in the operating point wasinvestigated using a test set of 40,800 transient stability simulations for 50 randomlygenerated operating points. Accuracy in excess of 95% was obtained for the 40,800contingencies.

One way to use DTs for real-time control is to train a DT to predict whether loss ofsynchronism will occur without control and train another DT to predict whether loss ofsynchronism will occur with some particular control. In simulations of a 176 bus modelof the western U.S., a combination of generator tripping at Palo Verde and load sheddingat Tesla and Vaca-Dixon was found to stabilize long duration three-phase faults for fivetransmission lines in the Arizona area [4-48]. A test set of 500 random duration, three-

4-18

phase faults on these lines without control contained 232 stable cases and 268 unstablecases. If control is applied when the DTs predict stable with control and unstable without,then 215 of the 232 stable cases have no unnecessary control intervention. The remaining17 stable cases had control intervention without adverse effect. The controller operated inall 268 of the unstable cases, and stabilized 263 of them. The remaining 5 cases had verylong fault durations and hence were too serious to control.

4.3.3 Decision trees for response-based controlPrior to 1996, the research on DTs for real-time control had assumed there would besome way to detect that an event had just occurred so that the immediate post-eventmeasurements could be fed into the decision tree. More recently, decision trees have beenadapted to continuously follow the measurements and select control action as soon as theneed becomes apparent [4-50,4-58]. This response-based operation effectively turns theclassifier approach into a natural generalization of the way engineers determine relaysettings and discrete-event control laws. For example, in the development of the R-Rdotout-of-step relay [4-56,4-57], apparent resistance R and its rate of change Rdot wereplotted for both stable and unstable transient events. The apparent resistance wasmeasured at Malin substation near the electrical center of the Pacific AC Intertie (PACI)in order to detect loss of synchronism across the PACI. Using large-scale simulations, thedesigners learned to differentiate between stable and unstable swings based on theirtrajectories in the R-Rdot phase plane. Decision boundaries were then drawn to classifynew swings as either stable or unstable and to order circuit breaker operation asappropriate.

Decision tree training algorithms can draw decision boundaries in phase planes as well asin higher dimensional spaces. The R-Rdot relay provides a good demonstration of DTsfor response-based control. Instead of using only the immediate post-event electricalmeasurements, response-based DT control is achieved by using every time sample in thesimulation for an input-output pair. Using 28,728 data points extracted from 168 transientsimulations on the 176 bus model, a DT was trained to associate each pair of R and Rdotmeasurements with whether the angle across the PACI exceeded 90 degrees when themeasurements were taken. The 168 contingencies in the training set contained 6 differentfault scenarios for each of 28 transmission lines: one-cycle fault, three cycle fault, fourcycle fault, six cycle fault, one cycle fault followed by loss of the Pacific DC Intertie(PDCI), and one cycle fault followed by loss of the Intermountain Power Project (IPP)DC line. All faults were three-phase short circuit to ground with the faulted line removedat clearing time. Each simulation in the training set was three seconds long. The test setcontained data extracted from 784 simulations which were five seconds long; 756 of thetest set events were double contingency outages, each involving two of the 28 study lines.The resulting DT tripped correctly on 70 events, tripped incorrectly on 10 events,correctly refrained from tripping on 704 events, and never failed to trip on an unstableevent. In addition to achieving response based control, these DTs also respondappropriately to single-phase faults. The training sets can be generated using industrystandard power system models.

4-19

Specifying misclassification costs during the training has been particularly helpful forbuilding DTs to perform response-based control. A circuit breaker controlled by this DTwill be programmed to trip and stay open once the DT outputs “trip.” Hence there is noremedy for a false trip; once the breaker opens it must stay open. If, however, the DT failsto trip on a case where the intertie angle has in fact exceeded 90 degrees, then it still hasthe option of tripping later. There will always be an area of uncertainty between when theDT should trip versus not trip. For a truly unstable event, the need to trip should becomemore obvious over time and it would be desirable to train the DT to wait until the need totrip is nearly certain. This behavior can be obtained by assigning a high misclassificationcost to false trips. The resulting DT will only trip if the trajectory enters a region wherestable trajectories almost never enter. For training the DT shown in Figure 4-11, themisclassification cost of false trips was set 50 times higher than the misclassification costof failures to trip.

R < 38

R < 0

No Trip Rdot < - 600

No Trip Rdot < - 64

R < 21

Trip Rdot < - 143

Trip No Trip

R < 17

Rdot < - 13

Trip No Trip

No Trip

No Trip

Yes No

Yes No

Yes No

Yes No

Yes No Yes No

Yes No Yes No

Fig. 4-11. Decision tree for an R-Rdot out-of-step relay.

4.3.4 Decision trees for improving dynamic performanceDecision trees can perform response-based discrete-event control to improve the dynamicperformance of stable transient events [4-49,4-50]. In order to automatically train aclassifier to associate the incoming measurements with an appropriate discrete-eventcontrol, it’s necessary for a computer algorithm to determine which control to assign eachcase in the training set. If a control makes the difference between stability and instability,then the choice is clear. When instability is not an issue and the goal is to improve thedynamic performance, an objective measure of the post-event behavior must be used. Thefollowing objective function is used to calculate the severity of simulated transient eventswith and without control.

dtMJ i

T

i i2

coa0)( δδ −= ∫ ∑

4-20

This performance index is like the weighted sum squared “error” comparing thesimulated swing curves to a hypothetical “ideal” trajectory where all the generator anglesare constant with no angle differences. The sum does not have to contain all thegenerators in the model. A sampling on the order of 10–100 of the larger generatorsdistributed throughout the power system is sufficient to have J be a fairly good numericalmeasure of the amount of interarea oscillation following a disturbance. Between twosimulations, everything is held fixed except for some control action that needs to beevaluated. Controls that reduce J tend to have the strongest smoothing and stabilizingeffects on the post-event oscillations. In addition to improving dynamic performance, theperformance index can also be used to determine powerful combinations of discrete eventcontrols for stabilizing strongly unstable events [4-48].

Decision trees were trained to improve dynamic performance using data extracted from93 transient simulations on the 176 bus model. Each contingency was simulated with andwithout a 500 MW fast power increase on the IPP DC line immediately after faultclearing, and a DT was trained to predict from real-time phasor measurements whetherthe numerical improvement in dynamic performance would exceed a threshold [4-50].The decision tree was tested on three cycle, three-phase faults and five cycle single line toground faults applied to the same 31 transmission lines used in the training set. The DTordered a 500 MW fast power increase at some point in 44 of the 62 simulations and hada positive effect in 42 of the 44 simulations it tried to control. Fifty-one of the 62simulations were stable for the first two seconds, and 39 of the 44 DT operations occurredduring stable events. The average performance index improvement for the 39 stablecontingencies was 2.4 and the maximum improvement was 4.7. Most of the stable eventshad performance index scores between 40 and 80. Using 60 as a rough estimate of theaverage score for stable cases, the improvement from the DT controller is roughly 2.4/60= 4.0%. For comparison, a 500 MW IPP DC ramp in response to the initial events wouldhave prevented the cascading outage that occurred on December 14, 1994 by reducingoverloads which caused some of the transmission lines to trip [4-49,4-50]. Performanceindex calculations applied to the large-scale simulations of the initial December 14 eventsshowed an improvement of 4.1% resulting from the DC fast power change.

References4-1 R. Fujiwara, T. Sakaguchi, Y Kohno, and H. Suzuki, “An Intelligent Load Flow

Engine for Power System Planning,” IEEE Transactions on Power Systems, Vol.PAS-3, pp. 302–307, August 1986.

4-2 C. C. Liu, H. Marathe, and K. Tomsovic, “A Voltage Control Expert System andIts Performance Evaluation,” in T. Dillon and M. Laughton, (eds.) Expert SystemApplications to Power Systems, Prentice-Hall, pp. 210–252, 1989.

4-3 CIGRÉ TF 38-06-02 on Expert Systems Applied to Monitoring and AlarmProcessing in Control Centers, T. Dillon (convener), “A Survey on ExpertSystems for Alarm Handling,” Electra, pp. 142–156, December 1992.

4-4 CIGRÉ TF 38-06-06 on Artificial Neural Network Applications for PowerSystems, Dagmar Niebur (convener), “Neural Network Applications in Power

4-21

Systems,” Int. Journal of Engineering Intelligent Systems, Vol. 1 No. 3, pp. 133–158, December 1993.

4-5 CIGRÉ TF38-06-06 on Artificial Neural Network Applications for PowerSystems, D. Niebur (convener), “Artificial Neural Networks for Power Systems,”Electra, No. 159, pp. 76–101, April 1995.

4-6 CIGRÉ TF 38-06-02 on Expert Systems Applied to Monitoring and AlarmProcessing in Control Centers, T.Dillon (convener), “Fault Diagnosis in ElectricPower Systems through AI Techniques,” Electra, No. 159, pp. 50–75, April 1995.

4-7 CIGRÉ TF 38-06-03, on Practical Applications of Expert System, “Summary ofFinal Report,” to appear in Electra.

4-8 L. A. Zadeh, “Fuzzy sets,”Inf. Control 8, pp. 338–353, 1965.

4-9 L.-W. Wang and J. M. Mendel, “Fuzzy Basis Functions, Universal Approximationand Orthogonal Least-Square Learning,” IEEE Transactions on Neural Networks,Vol. 3, No. 5, pp. 807–814, September 1992.

4-10 K. Lee, “Stability of Fuzzy Controllers,” in K. Tomsovic and M. Y. Chow (eds.),Tutorial on Fuzzy Logic Applications to Power Systems, tutorial publication of the1997 IEEE Power Industry Computer Applications Conference, Columbus, OH,May 11–15, 1997.

4-11 D. Niebur and O. Malik, “Control Applications of Fuzzy Systems,” in K.Tomsovic and M. Y. Chow (eds.), Tutorial on Fuzzy Logic Applications to PowerSystems, tutorial publication of the 1997 IEEE Power Industry ComputerApplications Conference, Columbus, OH, May 11–15, 1997.

4-12 M. A. M. Hassan, O. P. Malik and G. S. Hope, “A Fuzzy Logic Based Stabilizerfor a Synchronous Machine,” IEEE Transactions on Energy Conversion, Vol. 6,No. 3, pp. 407–413, September 1991.

4-13 T. Hiyama, M. Kugimiya and H. Satoh, “Advanced PID type Fuzzy Logic PowerSystem Stabilizer,” IEEE Transactions on Energy Conversion, Vol. 9, No. 3, pp.514–520, September1994.

4-14 T. Hiyama, S. Oniki, and H. Nagashima, “Evaluation of Advanced Fuzzy LogicPSS on Analog Network Simulator and Actual Installation on Hydro Generators”,IEEE Transactions on Energy Conversion, Vol. 11, No. 1, pp.125–131, March1996.

4-15 Y. M. Park, U. C. Moon, and K. Y. Lee, “A Self-Organizing Power SystemStabilizer Using Fuzzy Auto-Regressive Moving Average (FARMA) Model,”IEEE Transactions on Energy Conversion, Vol. 11, No. 2, pp. 442–448, June1996.

4-16 R. Fischl, D. Niebur,. and M. El-Sharkawi, “Applications of Artificial NeuralNetworks to Power System Security Assessment,” in M. El-Sharkawi. and D.

4-22

Niebur (eds.) Tutorial Course on Applications of Artificial Neural Networks toPower Systems, IEEE Catalog Number 96 TP 112-0, 104–127.

4-17 D. Novosel and R. King, “Using Artificial Neural Networks for Load Shedding toAlleviate Overloaded Lines,” IEEE Transactions on Power Delivery, Vol. 9, No.11, pp. 425–433, January 1994.

4-18 A. Piras, A. Germond, B. Buchenel, K. Imhof, and Y. Jaccard, “HeterogeneousArtificial Neural Networks for Short Term Load Forecasting,” Proceedings ofIEEE PICA ‘95, Salt Lake City, Utah, 1995.

4-19 D. J. Sobajic and Y.-H. Pao, “Artificial Neural-Net Based Dynamic SecurityAssessment for Electric Power Systems,” IEEE Transactions on Power Systems,Vol. 4, No. 4, 220–228, February 1989.

4-20 D. Niebur and A. J. Germond, “Power System Static Security Assessment Usingthe Kohonen Neural Network Classifier,” IEEE Transactions on Power Systems,Vol. 7, No. 2, 865–872, May 1992.

4-21 D. J. Sobajic, Y.-H. Pao, and J. Dolce “Real-time Security Monitoring of ElectricPower Systems Using Parallel Associative Memories,” IEEE InternationalSymposium on Circuit and Systems, New Orleans, Louisana, May 1–3, 1990, pp.2929-2932.

4-22 S. Weerasooriya and M. A. El-Sharkawi, “Use of Karhunen-Loève Expansion inTraining Neural Networks for Static Security Assessment,” First InternationalForum on Applications of Neural Networks to Power Systems, Seattle WA, pp.59–64, July 1991.

4-23 O. Saito, K. Koizumi, M. Udo, M. Sato, H. Mukae, T. and Tsuji, “SecurityMonitoring Systems Including Fast Transient Stability Studies,” IEEETransactions on Power Apparatus and Systems, Vol. 94, No. 5, pp. 1789–1805,September/October 1975.

4-24 D. J. Sobajic and Y.-H. Pao, “Artificial Neural-Net Based Dynamic SecurityAssessment for Electric Power Systems,” IEEE Transactions on Power Systems,Vol. 4, No. 4, pp. 220–228, February 1989.

4-25 M. Aggoune, M. A. El-Sharkawi, D. C. Park, M. J. Damborg, and R. J. Marks II,“Preliminary Results on Using Artificial Neural Networks for SecurityAssessment,” IEEE Proceedings of 1989 PICA, Seattle, Washington, U.S.A., pp.252–258, May 1989.

4-26 T. Dillon and D. Niebur (eds.) Neural Net Applications in Power Systems, CRLPublishing Ltd., Leics, UK, 1996.

4-27 T. Hiyama, S. Kouzuma, and T. Imakubo, “Evaluation of Neural Network BasedReal Time Maximum Power Tracking Controller for PV System,” IEEETransactions on Energy Conversion, Vol. 10, No. 3, pp. 543–548, September1995.

4-23

4-28 J.-A. Peças-Lopez, et al., “Dynamic Security Assessment using PatternRecognition, Neural Networks and Decision Trees - Results in the Lemnos PowerSystem,” Proc. of the Conference on Rough Sets and Soft Computing (RSSC'94),November 1994.

4-29 P. Shanahan and S. Naumann, “Evaluation of Simultaneous TransferCapabilities,” Proceedings of the 1995 American Power Conference, Chicago, IL,pp. 1463–1468, April 1995.

4-30 J. Kanetkar and S. Ranade, “Compact Representation of Power System Security—a review,” Proceedings of the North American Power Symposium, 1992.

4-31 R. Farmer, “Present Day Power System Stability Analysis Methods in the WesternUnited States,” Proceedings of the International Symposium on Power SystemStability, Ames, Iowa, pp. 39–44, May 13–15, 1985.

4-32 J. D. McCalley, B. A. Krause, “Rapid Transmission Capacity MarginDetermination for Dynamic Security Assessment using Artificial NeuralNetworks,” Electric Power Systems Research, Vol. 34, pp. 37–45, 1995.

4-33 J. D. McCalley, S. Wang, Q. Zhao, G. Zhou, R. T. Treinen, and A.Papalexopoulos, “Security boundary visualization for systems operation,” IEEETransactions on Power Systems, Vol. 12, No. 2, pp. 940–947, May 1997.

4-34 Y. Jacquemart, L. Wehenkel, T. Van Cutsem, and P. Pruvot, “StatisticalApproaches to Dynamic Security Assessment: The Database GenerationProblem,” Proceedings of IFAC Symposium on Control of Power Plants andPower Systems, Cancun, Mexico, December 1995.

4-35 V. Van Acker, S. Wang, J. C. McCalley, G. Zhou, and M. Mitchell, “DataGeneration using Automated Security Assessment for Neural Network Training,”Proceedings of the 29th North American Power Symposium, Laramie, WY,October 1997.

4-36 W. T. Miller, R. S. Sutton and P. J. Werbos, Neural Networks for Control, TheMIT Press, pp. 28–65, 1990.

4-37 P. J. Werbos, “Backpropagation Through Time: What it does and how to do it,”Proceedings of IEEE, pp.1550–1560, Vol. 78, No. 10, October 1990.

4-38 K. S. Narendra and K. Parthasarathy, “Identification and Control of DynamicSystems Using Neural Networks,” IEEE Transactions on Neural Networks, Vol.1, pp. 4-28, March 1990.

4-39 T. Yamada and T. Yabuta, “Neuro-controller Using Autotuning Method forNonlinear Functions,” IEEE Transactions on Neural Networks, Vol. 3, pp. 595–601, July 1992.

4-40 C. C. Ku and K. Y. Lee, “Diagonal Recurrent Neural Network for DynamicSystem Control”, IEEE Transactions on Neural Networks, Vol.6, pp. 144–156,January 1995.

4-24

4-41 M. Park, M. S. Choi and K. Y. Lee, “An optimal tracking neuro-controller fornonlinear dynamic systems,” IEEE Transactions on Neural Networks, Vol. 7, No.5, pp. 1099–1109, September. 1996.

4-42 Stagg and El-Abiad, Computer Methods in Power System Analysis, McGraw-Hill,pp. 387, 1968.

4-43 T. E. Kostyniack, PSS/E Program Operation Manual, Power Technology Inc.,October 31, 1983.

4-44 L. Breiman et al., Classification and Regression Trees, Wadsworth, Belmont,California, 1984.

4-45 J. H. Friedman, “A Recursive Partitioning Decision Rule for NonparametricClassification,” IEEE Transactions on Computers, C-26, pp. 404–408, 1977.

4-46 S. M. Rovnyak, S. E. Kretsinger, J. S. Thorp, and D. E. Brown, “Decision Treesfor Real-Time Transient Stability Prediction,” IEEE Transactions on PowerSystems, Vol. 9, No. 3, pp.1417–1426, August 1994.

4-47 S. E. Kretsinger, S. M. Rovnyak, D. E. Brown, and J. S. Thorp, “Parallel DecisionTrees for Predicting Groups of Unstable Generators from Synchronized PhasorMeasurements,” Precise Measurements in Power Systems Conference, Arlington,Virginia, October 25–29, 1993.

4-48 S. M. Rovnyak, C. W. Taylor, and J. S. Thorp, “Performance Index and ClassifierApproaches to Real-Time, Discrete-Event Control,” Control EngineeringPractice, Vol. 5, No. 1, pp. 91–99, 1997.

4-49 S. M. Rovnyak, C. W. Taylor, J. R. Mechenbier, and J. S. Thorp, “Plans toDemonstrate Decision Tree Control Using Phasor Measurements for HVDC FastPower Changes,” Fault and Disturbance Analysis & Precise Measurements inPower Systems Conference, Arlington, Virginia, Nov. 8–10, 1995.

4-50 Synchronized Phasor Measurements for the Western Systems CoordinatingCouncil, EPRI Final Report TR-107908, May 1997.

4-51 L. Wehenkel, Th. Van Cutsem, and M. Ribbens-Pavella, “An ArtificialIntelligence Framework for On-Line Transient Stability Assessment of PowerSystems,” IEEE Transactions on Power Systems, Vol. 4, No. 2, pp. 789–800,November 1989.

4-52 L. Wehenkel, M. Pavella, “Decision Trees and Transient Stability of ElectricPower Systems,” Automatica, Vol. 27, No. 1, pp. 115–134, 1991.

4-53 L. Wehenkel, M. Pavella, E. Euxibie, and B. Heilbronn, “Decision Tree BasedTransient Stability Method - A Case Study,” IEEE Transactions on PowerSystems, Vol. PWRS-9, No. 1, pp. 459-469, February 1994.

4-54 Y. Mansour, E. Vaahedi, A. Y. Chang, B. R. Corns, B. W. Garrett, K. Demaree,T. Athay, and K. Cheung, “B. C. Hydro’s On-line Transient Stability Assessment

4-25

(TSA) Model Development, Analysis, and Post-Processing,” IEEE Transactionson Power Systems, Vol. 10, No. 1, pp. 241–253, February 1995.

4-55 H. Ota, Y. Kitayama, H. Ito, N. Fukushima, K. Omata, K. Morita, and Y. Kokai,“Development of Transient Stability Control System (TSC System) Based on On-Line Stability Calculation,” IEEE Transactions on Power Systems, Vol. 11, No. 3,pp. 1463–1472, August 1996.

4-56 C. W. Taylor, J. M. Haner, L. A. Hill, W. A. Mittelstadt, and R. L. Cresap, “ANew Out-of-Step Relay With Rate of Change of Apparent ResistanceAugmentation,” IEEE Transactions on Power Apparatus and Systems, Vol. PAS-102, No. 3, pp. 631–639, March 1983.

4-57 J. M. Haner, T. D. Laughlin, and C. W. Taylor, “Experience with the R-Rdot Out-of-Step Relay,” IEEE Transactions on Power Delivery, Vol. PWRD-1, No. 2, pp.35–39, April 1986.

4-58 S.M. Rovnyak and Y. Sheng, “Using Measurements and Decision Tree Processingfor Response-Based Discrete-Event Control,” Proceedings of IEEE/PES 1999Summer Meeting, pp. 10-15, Edmonton, July 18–22, 1999.

4-59 D. Y. Abramovitch and L. G. Bushnell, “Report on the Fuzzy versus ConventionalControl Debate,” IEEE Control Systems, pp. 88–91, July 1999 (debate betweenMichael Athans and Lotfi Zadeh).

4-60 P. P. Bonissone, et al., “Industrial Applications of Fuzzy Logic at GeneralElectric,” Proceedings of the IEEE, Vol. 83, No. 3, pp. 450–465, March 1995.

4-61 S. Chiu, “Using Fuzzy Logic in Control Applications: Beyond Fuzzy PIDControl,” IEEE Control Systems, pp. 100–104, October 1998.

Chapter 5

Integration of Dynamic Security Assessmentand Stability Controls

Dynamic security assessment (DSA) or transient security assessment (TSA) determines asystem’s ability to survive contingencies with a safety factor (margin). To ensure that a systemremains dynamically secure, preventive or corrective remedial actions are designed. Preventiveactions are applied in the pre-contingency system so that after any credible contingency thesystem remains secure. Examples include restrictions on interface flows, angle differences acrossa particular interface, and total generation out of a plant. Corrective remedial actions (stabilitycontrols) are those taken following a contingency. Examples include generator or load tripping,and capacitor bank or reactor switching.

Traditionally, preventive and corrective dynamic security measures have been developed fromnumerous off-line simulations. Transfer limits are determined by selecting extreme systemconditions and simulating critical contingencies. The limits derived are conservative, since theyare based on extreme system conditions.

Recently, on-line dynamic security assessment tools have been developed [5-1–7,5-19,5-21] afew of which have found their way to real system implementation [5-1–3,5-6,5-7]. These toolsdiffer in the methodology but they share the same concepts and fundamental blocks. This chapterdescribes on-line dynamic security assessment methods as part of the remedial stability controldetermination, and describes in detail its different components. Transient security assessment forarming of generation tripping stability control is described. Stability control is made adaptivebased on the on-line security assessment.

5.1 On-Line Transient Stability Assessment DesignAn on-line DSA tool should meet the following requirements:

• Reliability. Both the hardware and software of on-line DSA should perform reliably under allfeasible system operating conditions.

• Accuracy. The accuracy of DSA is of ultimate importance to ensure the dynamic security of apower system. In particular, it should accommodate detailed models for system components,as well as for disturbances that may include autoreclosures and other complex switchingactions. The general trend is that on-line DSA should give comparable results with the bestoff-line study tools for a given system model.

• Performance. The processing speed of DSA is often critical in meeting the requirements foron-line real-time or near real-time operations. In order to achieve the best computation speed,advanced techniques in software and hardware design must be used. Other performancerequirements for DSA include flexibility in data input/output and good user interface.

As described in the following subsections, on-line dynamic security assessment consists of thefollowing four elements:

• preprocessing

5-2

• security assessment

• post-processing

• process control and integration

5.1.1 PreprocessingThe task of preprocessing includes static state estimation, housekeeping for base casedevelopment, and contingency screening and ranking.

A major impediment to DSA for interarea stability problems in large interconnections is thedifficulty of state estimation to obtain the on-line power flow base case. For a particular controlcenter, the main difficulty is with the external network model. Considerable inter-utility dataexchange is required. There are other difficulties associated with measurement accuracy,unbalance operation, network parameter uncertainty, etc. [5-24].

State estimation can be improved by high quality digital measurements from throughout theinterconnection. Synchronized positive sequence phasor measurements are valuable [5-22,5-25,5-26]. Conceptually, with high quality bus voltage magnitude and angle measurements, buspower flow states are known.

External network models for on-line DSA is obtained by selecting from a number of previouslystored dynamically reduced system models [5-1,5-2]. Alternatively dynamic reduction techniquescan be used in real-time to develop the external model. This will facilitate base case initializationand helps maintain the base case within certain size limit so that the computation speedrequirements can be met.

In addition to power flow data, other data required for DSA may also need to be updated for anew system snapshot. For instance, the settings of a PSS for a pumped storage generation unitmay need adjustment for the different modes of operation of the unit. The contingencies may alsoneed update when the network topology or system operation condition changes.

It’s impossible to assess all the credible contingencies within the confines of availablecomputational resources and required response times. Therefore, the list of crediblecontingencies has to be reduced to make it manageable by the security assessment module.

The contingency screening and ranking method could be based on transient energy functions[5-10–17], expert systems [5-2,5-8,5-30], neural networks [5-8,5-9], extended equal area criteria[5-17–20], or indices derived from energy properties or fast time domain simulations usingsimplified models [5-6,5-7,5-10]. The common requirements of candidate contingency screeningand ranking methods are high speed and accuracy of the final results. While all these methodscan be used for contingency screening and ranking, the final limit computations should be doneusing more accurate methods.

The performance of a contingency screening and ranking method can be evaluated in terms of itsmisclassification. Misclassification consists of two components of False Alarm and FalseDismissal as described below.

• False alarm: a stable contingency that is identified as an unstable one (critical one)

5-3

• False Dismissal: an unstable contingency that is identified as a stable one (non-critical one)

An acceptable contingency screening and ranking method should have zero false dismissals anda very low number of false alarms.

5.1.2 Security assessmentDetailed time domain simulation is the most reliable security assessment approach [5-1–7,5-19,5-21]. Commercial-grade software is available that can normally be customized for a utility’srequired modeling and disturbances. This allows on-line security assessment with unlimitedmodeling capabilities capable of handling a full-scaled power system.

Normally, DSA assesses transient stability of a power system, or the ability of the system tomaintain synchronism after a credible contingency. As power systems operate in more and morestressed conditions, another form of angle stability, i.e., small-signal stability in the form ofsustained or growing oscillations in part or all of the system, may become critically restricting tothe system operating limits. This has already happened in some parts of the North Americaninterconnected power systems. The requirement to address this type of stability problem calls foran efficient and reliable method to compute the critical mode of the system. This is still an areawith room for research. One approach based on the time domain simulation technique is to obtainan estimate of the critical mode by post-processing simulation results. This is further described inthe following section.

5.1.3 Post-processingOnline implementation of time simulation requires a built-in intelligence for the following:

• Assessing the system dynamic performance (stable, unstable).

• Determining the degree of stability or instability (margin).

• Determining the sensitivity of the margin to key variables (transfer limit and generationtripping).

• Determining the transfer limit or preplanned stability control actions (e.g., arming ofgenerator tripping).

In the following, several methods that have been used in the post-processing stage are described.

Second-kick method [5-2, 5-4]. The second kick method was based on energy concepts fordetermining stability margin and other useful information from the simulations. It was inspiredby the hybrid method [5-3]. Although there are different implementation methods available forthis algorithm [5-5,5-29], the original concept [5-2] is described below.

Detailed time domain simulation is performed with calculation of potential energy, kineticenergy and corrected kinetic energy. No modeling assumptions are made and no analyticalequation is used to calculate potential energy. The minimum of the corrected kinetic energy,Kemin1 is identified, after the contingency. If the minimum is greater than zero (systemunstable), the margin is calculated from the value of the corrected kinetic energy at this point. Ifthe system is stable (corrected Kemin = 0), at Tkemin1, a second fault (second kick) which is

5-4

long enough to make the system unstable, is applied and simulation is continued until the secondminimum of kinetic energy, Kemin2, is obtained. This point also reflects the crossing of thepotential energy boundary surface (PEBS), as shown in Figure 5-1. The transient energy marginis then calculated using the values of the corrected kinetic energy at the second minimum ofkinetic energy (Kemin2) and the value after the second fault recovery (Kerec2) taking intoaccount adjustments due to potential energy change during the second kick. Figure 5-1 shows thesystem trajectory on the potential energy surface.

The basic idea here is that the kinetic energy injected into the system by the second kick minusthe value of the kinetic energy left in the system at the crest of potential energy hill, (PEBScrossing) should give the transient energy margin. This value should be adjusted for the potentialenergy change during the second kick. The transient energy margin, therefore is calculated by:

TEM = Kerec2- Kemin2 + Dpe

where Dpe is the change in potential energy during the second kick.

Fig. 5-1. Corrected kinetic variations.

Extended equal area criterion (EEAC) [5-17, 5-18]. The EEAC was developed based on thefact that the loss of synchronism in a power system is always initiated from the splitting of thesystem into the following two parts:

• Critical cluster of generators (CCG)

• Rest of the system

These methods have evolved in their developments through three major stages, each of which ischaracterized by a special version of the method:

• Static method

5-5

• Dynamic method

• Integrated method

The SIngle Machine Equivalent (SIME) method belongs to the last type [5-21].

The basic difference among these versions is the number of Critical Cluster Center of Inertia(CCCOI) transformations that are performed to obtain the parametric One-Machine Infinite Bus(OMIB) system. The static method does only one static transformation and therefore its accuracyis usually not satisfactory. The dynamic method improves the accuracy by using severaltransformations. This is achieved by simplifying power system modeling, and by using theTaylor-series expansion technique to obtain the approximate trajectory of the system. In theintegrated method, the transformation is integrated with the detailed time-domain simulations.Thus, no modeling compromise is required and the stability index so computed is very accurate.

Figure 5-2 shows the principle of the integrated EEAC. System snapshots are taken from theconventional time-domain simulation results (Figure 5-2 (a)) and for each snapshot a CCCOItransformation is performed to obtain the parametric OMIB system trajectory (Figure 5-2 (b)).The stability index η of the system can then be defined as

Thus, -100 ≤ η ≤ 100, and

η ≤ 0 if the system is unstable

η > 0 if the system is stable

The computation of this index requires a straightforward implementation of the integratedmethod on top of the time-domain simulation engine. As described earlier, such an index is notsubject to any modeling restrictions and it is also able to identify multi-swing stability problem.

In systems for which angle stability is the only concern to the dynamic security, the integratedmethod implementation can be made to check the system status during the simulation. If thesystem is found to be definitely stable or unstable prior to the end of the simulation, thesimulation is terminated. For all other cases a complete simulation is required. This is an area forwhich more research is needed to equip this class of methods with more sophisticated earlytermination techniques.

>−×

>−×=η

)AA(unstableissystemtheIfA

AA100

)AA(stableissystemtheIfA

AA100

decincinc

incdec

incdecdec

incdec

5-6

Fig. 5-2. Integrated EEAC.

Other methods to determine stability margin. Reference 5-6 defines a stability index using dotproducts of generator rotor angles, speeds, and accelerating powers obtained from time-domainsimulations. The idea is based on the path of the post-fault system trajectory: the system is stableif its trajectory “swings back” before reaching PEBS or unstable if its trajectory “exits” thePEBS. A DSA system using this index has been developed and is now operational on-line.

Reference 5-7 describes an approach for establishing the required generator tripping. It uses onetime domain simulation and with the help of the generator’s angular swing and kinetic energyestimates the required tripping. Using a conservative threshold, the authors have verified theperformance of the method on a developed prototype. The “Transient Stability Control” systemhas been in service since June 1995, and has since been extended.

Reference 5-31 uses the signal energy obtained from time-domain simulations to define astability index. The authors successfully applied this index to determine the transient stabilitytransfer limit for the Hydro Quebec system.

Stability Limit Calculation: The final outcome of TSA are guidelines for system operation inthe form of pre-contingency transfer limits and generation tripping remedial action immediatelyfollowing a fault.

Several approaches have proposed to determine the power transfer limits, all of which make useof the stability margin or index as a measure of system stability:

δ

P

Pm

Pe

Ainc

Adec

t

δ

(b) Parametric OMIB system trajectory(Power-angle characteristic)

Pe Electric powerPm Mechanical powerAdec Kinetic energy decreasing areaAinc Kinetic energy increasing area

(a) Multi-machine system trajectory(Time-domain)

System snapshot

CCCOI transformation

5-7

• Sensitivity-based method [5-5,5-20].

• Curve fitting method [5-6].

• Binary and accelerated binary search method [5-31].

Figure 5-3 shows the application of the sensitivity-based method (in this case, the stabilitymargin obtain by the second kick approach is used). To find the power transfer limit and therequired generation tripping, sensitivity values are calculated for stability margin with respect togeneration tripping, or generation change in the case of power transfer limit calculations. Theseanalytical equations can only be used in the first step to calculate the conditions for the next oneand be abandoned afterwards. After the second run linear interpolation is used to obtain thesensitivity values from the two previous stability margin calculations.

0

1

-1 Iteration 1

Iteration 2 Iteration 0

100 200

Generation Rejection , MW

Energy Margin

Fig. 5-3. Limit calculation using sensitivity factors.

Historically, utilities have established remedial action controls, e.g. units for generationshedding, based on design limitations and experience. There seems to be a need for rigorousmethods which can establish the most effective remedial actions in real-time. Reference 5-28 hasmoved in that direction by establishing the corrective actions necessary to stabilize all dangerouscontingencies simultaneously, while ensuring the maximum allowable transfer between areas.

Critical damping estimate: As mentioned earlier, an increasing concern on the angle stability ofpower systems is the oscillation problems, and DSA needs to handle these problems. Reference5-32 presents a method of estimating the damping of the critical mode (i.e., the least stablemode) in a system by using the multi-channel Prony analysis. The advantage of this approach isthat it makes use of the time-domain simulation results with very small computational overhead.The multi-channel Prony algorithm helps improve drastically the accuracy of the results ascompared with the commonly used single-channel analysis.

This method is illustrated in the following, using a system model consisting of 6139 buses and798 generators. Figure 5-4 shows the simulation results for selected generator rotor angles for atypical contingency. It can be seen that, although the system is transiently stable, sustainedoscillations exist. It’s therefore important to identify the critical mode in order to assess thesecurity of the system.

5-8

Table 5-1 contains the critical mode identified using a four-channel Prony analysis algorithm. Forcomparison, the results from the full eigen-value analysis and the conventional Prony analysis oneach individual channel are also shown in the table. The eigen-value analysis clearly shows thecritical mode at 0.79 Hz with almost zero damping; so does the four-channel Prony algorithm.However, the results from individual generator rotor angles are apparently not reliable.

Fig. 5-4. Simulation results for the sample system.

Table 5-1: Critical mode comparison from different methods

Computation Method Freq. (Hz) Damping (%)Prony on rotor angle of generator ‘A’ 0.71 2.57

Prony on rotor angle of generator ‘B’ 0.80 -3.20

Prony on rotor angle of generator ‘C’ 0.62 -3.75

Prony on rotor angle of generator ‘D’ 0.79 -3.11

Four-channel Prony on rotor angles of ‘A’, ‘B’, ‘C’, ‘D’ 0.75 -0.58

Full eigen-value analysis 0.79 0.03

5.1.4 DSA performance enhancementsAs an on-line application, the computation speed of DSA has been of critical importance to theend users. As typical requirements, an on-line DSA system should be able to process hundreds ofcontingencies for dozens of transactions within a 10 to 20 minutes cycle. Further complicatingthe matter is the tendency that utilities are using larger and larger base case models in their EMS.It is inevitable that on-line DSA systems have to work with EMS models of 5,000 to 10,000

Gen

era

tor a

ngl

e (d

egre

es)

Time in seconds0 1 2 3 4 5

-100

-50

0

50

100

150

5-9

buses and up to 1,000 generators. Thus, techniques that speed up DSA performance need to bedeveloped and deployed in order to meet the requirements.

Among the techniques that aim at improving DSA performances, the following are noteworthy:

• Development of better contingency screening method. Any fast contingency screeningmethod that can reduce the false alarm rate while maintaining zero false dismissals candrastically reduce the computation time needed for detailed analysis of the criticalcontingencies.

• Enhancement of Early termination techniques. When using time-domain simulations fordetailed analysis of the critical contingencies, it’s desirable to terminate the simulation assoon as the stability of the system can be identified using a stability index.

• Parallel or distributed computations. This is the classical method of improving the speed ofsimulations. Since DSA involves multi-transaction, multi-contingency simulations, parallelor distributed simulations for transactions or contingencies can be easily achieved.

5.2 Other Integration of DSA and Stability ControlsThe previous section described DSA methods where the output is used for stability controladaptation—namely, arming of the correct number of generating units for tripping. This can bethought of as very slow, outer-loop adaptive supervisory control.

Another potential use of DSA in advanced stability control is pattern recognition based controlwhere DSA provides the database. This is described in Chapter 4.

Other synergism is possible between DSA and stability controls. High quality digitalmeasurements can both improve state estimation as described above, and be used for directmonitoring and stability control. Synchronized positive sequence phasor measurements are onetype of digital measurements. Phasor measurements may be sufficiently related to dynamic statessuch as rotor angles and speeds to be useful for stability control; see discussion in §4.3. Inaddition, fast digital measurements support stability control development, commissioning, andmonitoring as discussed in Chapter 2.

Another application of DSA for stability controls using measurements is found in reference [5-27]. In this work, the real-time transient stability emergency controls are derived by feeding theSingle-Machine Equivalent method [SIME] with real-time measurements taken at the powerplants to control the system transient stability in real-time and in a closed-loop fashion. The mainsteps of this approach are the prediction (say, 150 to 200 ms ahead) of the transient stabilitystatus of a system after a fault occurrence and its clearance by protective relays and its degree ofinstability if instability is detected. In the latter case, the amount of generation tripping requiredto compensate for this margin is assessed and the system status after the corrective action hasbeen triggered is monitored to establish whether this action is sufficient or additional remedialaction is required.

Complementing the computer-based contingency analysis described in the previous section,monitor-based DSA is valuable for both system operators, and for stability control analysts anddevelopers [5-23].

5-10

References5-1 K. Demaree, T. Athay, K. Cheung, Y. Mansour, E. Vaahedi, A. Chang, and B. Corns,

“An On-line Dynamic Security Analysis System Implementation,” IEEE Transactions onPower Systems, Vol. 9, No. 4, pp. 1716–1722, November 1994.

5-2 Y. Mansour, E. Vaahedi, A. Chang, B. Corns, B. Garrett, K. Demaree, T. Athay, and K.Cheung, “B.C. Hydro’s On-line Transient Stability Assessment (TSA) ModelDevelopment, Analysis, and Post-processing,” IEEE Transactions on Power Systems,Vol. 10, No. 1, pp. 241–253, February 1995.

5-3 G. A. Maria, C. Tang, and J. Kim, “Hybrid Transient Stability Analysis,” IEEETransactions on Power Systems, Vol. 5, No. 2, pp. 384–393, May 1990.

5-4 C. K. Tang, C. E. Graham, M. El-Kady, and R. T. H. Alden, “Transient Stability Indexfrom Conventional Time Domain Simulation,” IEEE Transactions on Power Systems, pp.1524–1530, August 1994.

5-5 E. Vaahedi, Y. Mansour, and Y. Tse, “A General Purpose Online Dynamic SecurityAssessment Method,” presented at the IEEE Summer Meeting, July 1997, Berlin,Germany.

5-6 G. C. Ejebe, C. Jing, J. G. Waight, V. Vittal, G. Pieper, F. Jamshidian, D. Sobajic, and P.Hirsch, “On-line Dynamic Security Assessment: Transient Energy Based Screening andMonitoring for Stability Limits,” EPRI Workshop on DSA/VSA, October 9–10, PaloAlto.

5-7 H. Ota, Y. Kitayama, H. Ito, K. Omata, K. Morita, and Y. Kokaki, “Development ofTransient Stability Control System (TSC) Based on On-line Stability Calculations,” IEEETransactions on Power Systems, Vol. 11, No. 3, pp. 1463–1472, August 1996.

5-8 A. B. R. Kumar, V. Brandwajan, and A. Ipakchi, Power System Dynamic SecurityAssessment Using Artificial Intelligence Systems, EPRI Final Report, RP3103-02, April1994.

5-9 Y. Mansour, E. Vaahedi, A. Chang, B. Corns, J. Tamby, and M. A. El-Sharkawi, “LargeScale Dynamic Security Screening and Ranking Using Neural Networks,” IEEETransactions on Power Systems, Vol. 12, No. 2, pp.954–962, May 1997.

5-10 S. Mokhtari, Analytical Methods for Contingency Selection and Ranking for DynamicSecurity Assessment, EPRI Final Report, RP3103-03, May 94.

5-11 T. Athay, R. Podmore, and S. Virmani, “A Practical Method for Direct Analysis ofTransient Stability,” IEEE Transactions on Power Systems, Vol. PAS-98, No. 2, pp. 573–584, March/April1979.

5-12 M. A. Pai, Energy Function Analysis for Power System Stability, Kluwer AcademicPublishers, 1989.

5-13 A. Fouad and V. Vittal, Power System Transient Stability Analysis Using TransientEnergy Function Method, Prentice Hall, 1990.

5-11

5-14 W. W. Price, “Rapid Analysis of Transient Stability,” IEEE Report No: 87TH0169-3-PWR, September 1987.

5-15 H. Chiang, C. Chu, and G. Cauley, “Direct Stability Analysis of Electric Power SystemsUsing Energy Functions: Theory, Applications, and Prospective,” Proceedings of IEEE,Vol. 83, No. 11, pp. 1497–1529, November 1995.

5-16 P. Baratella, B. Cova, M. Damonte, E. Gaglioti, R. Marconato, P. Scarpellini, “FastSimulation of Power System Dynamic in General-purpose Simulator,” ControlEngineering Practice, Vol. 5, No. 1, pp. 123–129, 1997.

5-17 A. Rahimi and G. Schaffer, “Power System Transient Stability Index for On-line Analysisof ‘Worst -Case’ Dynamic Contingencies,” IEEE Transactions on Power Systems, Vol.PWRS-2, No.3, pp. 660–668, August 1987.

5-18 Y. Xue and M. Pavella, “Extended Equal-Area Criterion: An Analytical Ultra-fastMethod for Transient Stability Assessment and Preventive Control of Power Systems,”International Journal of Electrical Power & Energy Systems, Vol. 11, No. 2, pp. 131–149, 1989.

5-19 Y. Xue, Y. Yu, J. Li, Z. Gao, C. Ding, F. Xue, L. Wang, G. K. Morison, and P. Kundur,“A New Tool for Dynamic Security Assessment of Power Systems,” IFAC/CIGRESymposium on Control of Power Systems and Power Plants, Beijing, China, pp. 604–609,1997 and also appeared in Control Engineering Practice, Vol. 6, pp. 1511–1516, 1998.

5-20 L. Wang, X. Wang, K. Morison, P. Kundur, F. Xue, C. Ding, Y. Luo, and Y. Xue,“Quantitative Search of Transient Stability Limits Using EEAC,” to be published in IEEEspecial publication on Techniques for Transient Stability Limit Searches.

5-21 Y. Zhang, L. Wehenkel, P. Rousseaux, and M. Pavella, “SIME : A Hybrid Approach toFast Transient Stability Assessment and Contingency Selection,” EPES, Vol. 19, No. 3,pp. 195–208, 1997.

5-22 I. W. Slutsker, S. Mokhtari, L. A. Jaques, J. M. Gonzalez Provost, M. B. Perez, J. B.Sierra, F. G. Gonzalez, and J. M. M. Figueroa, “Implementation of Phasor Measurementsin State Estimator at Sevillana de Electricidad,” IEEE/PES Power Industry ComputerApplications Conference, Salt Lake City, Utah, May 7–12, 1995.

5-23 J. Hauer, D. Trudnowski, G. Rogers, W. Mittelstadt, W. Litzenberger, and J. Johnson,“Keeping an Eye on Power System Dynamics,” IEEE Computer Applications in Power,Vol. 10, No. 1, pp. 26–30, January 1997.

5-24 M. M. Adibi and R. J. Kafka, “Minimization of Uncertainties in Analog Measurementsfor Use in State Estimation,” IEEE Transactions on Power Systems, Vol. 5, No.3, pp.902–910, August 1990.

5-25 A. G. Phadke, J. S. Thorp, and K. J. Karimi, “State Estimation with PhasorMeasurement,” IEEE Transactions on Power Systems, Vol. PWRS-1, No. 1, pp. 233–241, February 1986.

5-12

5-26 J. S. Thorp, A. G. Phadke, and K. J. Karimi, “Real Time Voltage-Phasor Measurementsfor Static State Estimation,” IEEE Transactions on Power Apparatus and Systems, Vol.PAS-104, No. 11, pp. 3098–3106, November 1985.

5-27 Y. Zhang, L. Wehenkel and M. Pavella, “A Method for Real-Time Transient StabilityEmergency Control,” Proc. of CPSPP’97, IFAC/CIGRE Symp. on Control of PowerSystems and Power Plants, Beijing, China, pp. 673–678, August 1997.

5-28 A. L. Bettiol, L. Wehenkel and M. Pavella, “Transient Stability-Constrained MaximumAllowable Transfer,” IEEE Transactions on Power Systems, Vol. 14, No. 2, pp. 654–659,May 1999.

5-29 K. W. Cheung, “A New Hybrid Method for On-line Dynamic Security Assessment,”IFAC Control Engineering Practice, 6 (1998), pp. 1373–1380.

5-30 K. W. Cheung, R. Paliza, T. K. Ma, T. Athay, J. Zuk, “An Expert System Guided On-lineDynamic Security Assessment System,” Proceeding of International Conference onIntelligent System Application to Power Systems (ISAP’94), Montpellier, France, pp. 263–270.

5-31 R. Marceau, M. Siraogue, S. Soumare, and X. D. Do, “Signal Energy Search Strategiesfor Transient Stability Transfer Limit Determination,” to be published in IEEE specialpublication on Techniques for Transient Stability Limit Searches.

5-32 M. Gibbard, N. Martins, J. J. Sanchez-Gasca, N. Uchida, V. Vittal, L. Wang, “RecentApplications Of Linear Analysis Techniques,” presented at the panel session at the 1998IEEE/PES Summer Meeting in San Diego.

Chapter 6

Measurement and Communication Technology

A stability controller addressing global objectives needs a reliable source of globalinformation. Local signals may prove inadequate for this, even when reinforced byextensive modeling studies. Fortunately, not all remote signals are equally vital tocontroller performance, or equally difficult to transmit reliably.

It’s useful to distinguish among:

• Modulation signals used to directly shape the controller output, u(t)

• System status flags used in

- rule-based control laws (e.g., parameter scheduling)

- coordination with other controls

- remote controller supervision

• Secondary response signals for

- direct testing of power system dynamics and controller effects

- local monitoring of controller performance

- alternate or supplemental modulation signals

Requiring that all modulation signals be local can make controller siting a difficultrobustness issue [6-1,6-2]. There are many aspects of the controller environment whichcannot be predicted from model studies, and which may not be measurable until thecontroller itself is available for system dynamics testing. Providing the controller (and thecontrol engineer) an ample reserve of directly measured dynamic information canincrease controller performance and robustness.

The Bonneville Power Administration (BPA), through long involvement in stabilitycontrol projects, has conducted numerous measurements of system dynamics [6-3–8].This has produced many examples of apparently anomalous system behavior. Most havebeen attributable to the power system itself. Some, however, may have involved falseoutputs from the transducers being used, or from the sensors (instrument transformers)that provide inputs to those transducers. Other likely sources include communicationchannels and secondary control loops, especially those in which modulation or samplingprocesses can translate signal components from one frequency to another.

False measurements will, at best, produce an erroneous view of the power system. Thiscan readily lead to inappropriate engineering or operational decisions. The situation ismore serious when false measurements enter the modulation loop of a major controlsystem (Figure 6-1). It’s likely that any extraneous signals emitting from the transducerwill be amplified by the actuator and re-injected into the power system. This is, at best, asource of undesirable noise disturbances. It’s also a potential path for disruptiveinteractions between the actuator and dynamic processes other than those targeted for

6-2

control. The trend toward fast power electronic actuators, together with more aggressivecontrol objectives, have sharply increased the risks in this respect.

CONTROL

LAW

command u(t)^

~nonlinear response u(t)

actuator noise υu(t)

controller input y m(t)

linear response u(t)–

load noise u L (t)

external inputs u E (t)unmeasured response y’(t)

input u(t)measured response y(t)

extraneoussignals

measurement

noise u m(t)

nonlinearinteractions

ACTUATOR

Sensors&

Transducers

POWER

SYSTEM

Fig. 6-1. Transducer effects in a closed loop controller.

Most stability controls seek to influence generator “swing” activity, or else the voltagesupport that the generators provide to the transmission network. The dynamics undercontrol rarely occur at bandwidths above 2 Hz. The actuator, and the control law drivingit, may very well have bandwidths higher than this. HVDC links and SVCs, for example,may have controllable responses up to 20–25 Hz. This encourages the use of “highspeed” transducers with bandwidths in this same frequency range, to obtain full actuatorperformance. It also mandates the use of such transducers in monitoring of controllerperformance.

The vast majority of the transducers now in service are analog devices with bandwidthsin the range of 1 to 2 Hz. Advanced designs can achieve bandwidths approaching 30 Hz,in principal at least. Enhanced analog transducers used at the BPA have a bandwidth of20 Hz.

Field experience suggests that the interpretation and use of outputs from these or anyother transducers should be approached carefully. Transducers of all types may be subjectto aliasing effects that might, for example, permit mechanical or network resonances tomimic swing dynamics. Increasing transducer bandwidth also increases the likelihoodthat such effects will be present in the transducer output. Table 6-1, based in part fromreference 6-11, indicates that there are many candidate sources for extraneous transduceroutputs.

Desired characteristics for next-generation transducers include:

• Rigorous protection against out-of-band input signals.

• Absence of processing artifacts, such as spurious outputs.

• Programmable outputs, for versatility of application.

6-3

• High resolution and bandwidth, in control applications.

• Assured high accuracy, in metering applications.

• Good networking options, both local and wide area.

• Option for synchronizing measurements against a precise external reference.

• Overall affordability, considering all cost elements.

Such transducers will almost certainly require digital technology.

Table 6-1.Extraneous dynamics in the transducer environment

Dynamic Activity Frequency Range -Hz

Torsional oscillations 5 – 120Transient torques 5 – 50 Turbine blade vibrations 80 – 250Fast bus transfer 1 – 1000Controller interactions 10 – 30 Harmonic interactions and resonances 60 – 600Ferroresonance 1 – 1000 Network resonances 10 – 300

6.1 Introduction to TransducersFor our immediate purposes a transducer is a signal processing device that translatesinstantaneous “point on wave” current and voltage signals into averaged measures ofelectrical behavior. Chief among these are rms (root mean square) voltage, rms current,rms power, waveform frequency, and relative angles for voltage and current. Suitablechoices among these measures—and for the averaging times used in calculating them—are determined by the information that is needed. Reference 6-12 provides an overview ofstandard transducer types.

Existing transducer technology reflects a broad range of information needs. The slow endof the spectrum is occupied by revenue meters, which sacrifice dynamic response forhigh reliability and accuracy. At the other extreme, a digital relay is designed to detectand assess dynamic events with no more accuracy than reliability demands. Transducersfor stability control occupy a broad middle range, contingent upon the:

• Kind of dynamic process to be stabilized. Possibilities include

- local swing modes (with modulation on individual generators)

- interarea swing modes (with modulation on HVDC or FACTS device)

- voltage dynamics

• Role of the transducer in the control process. Possibilities include

- modulation signal for feedback control

6-4

- monitoring of power system conditions and behavior

- monitoring of controller activity, especially anomalous interactions

The transducer for a feedback modulation system would, by choice, be equipped withinternal and external filters protecting the feedback loop from interactions withextraneous dynamics. By contrast, a transducer for monitoring controller performanceshould be capable of detecting such extraneous dynamic activity. It would probably bemore lightly filtered, and it might have a bandwidth approaching 25 or 30 Hz. Ifinteractions are sensed at such frequencies it is highly desirable that direct waveformrecordings be made on a local digital fault recorder (DFR) or similar device. Figure 6-2provides an example of current waveforms under moderately disturbed conditions [6-13].

A Phase

B Phase

C Phase

Fig. 6-2. Measured current at Olinda substation for COTP Test Fault #3 (1715 h on03/23/93).

168 170 172 174 176 178-200

-150

-100

-50

0

50

100

150

Time in Seconds

Malin-Round Mountain #1 MW : PG&E Malin Sum MW : PG&E Olinda MW

Malin-Round Mtn1 MW swing

PG&E Malin Sum MW/2

PG&E Olinda MW

Fig. 6-3. Time response of Malin area transducers, insertion of Chief Joseph dynamicbrake on August 10, 1996.

6-5

Figures 6-3 through 6-5 indicate the relative performance of three kinds of transducers, asobserved through the microwave channels that communicate their outputs to BPA’scontrol center. All three transducers measure components of the real power export toPacific Gas & Electric (PG&E) on the California – Oregon Interconnection, or COI. Thesignal for Malin – Round Mountain circuit 1 is taken from an enhanced analog transducerthat has a bandwidth of 12–14 Hz, and that communicates on a 20 Hz channel. At theother extreme, the transducer for the PG&E – Olinda exchange is a conventional analogtransducer communicating through a low-bandwidth channel (probably 1.5 Hz or lower).

Differences among the signals in Figure 6-3 are almost entirely due to theinstrumentation. Ignoring transducer/channel dynamics, the PG&E – Malin signal shouldbe very close to twice that for Malin – Round Mountain circuit 1, and the PG&E – Olindasignal should be very similar except for magnitude. It is apparent that much of thewaveform detail is not tracked very well by the slower instrumentation, and that thewaveforms also exhibit appreciable delays.

Quantitative measures of relative response can be obtained by correlating the transducersignals against one another. Figure 6-4, based upon ambient noise outputs, compares theslowest signal against the fastest. Since the fast transducer has a much higher bandwidth(e.g., its response is nearly flat to well past 1 Hz), it appears that the much slow Olindatransducer has a -3 dB bandwidth near 0.9 Hz. Its relative time delay, evident in the linearphase characteristic which produces a lag of 180 degrees near 0.85 Hz, can be estimatedas (180)/(360*0.85) = 0.59 second. This value is consistent with Figure 6-3, but moreaccurate than what would be obtained by direct inspection.

0 1 2-40

-30

-20

-10

0

PG&E Olinda MW relative to Malin-Round Mountain #1 MW

Frequency in Hertz

-180

-90

-60

0

60

90

180

GAIN

PHASE

-3 dB

TF

Gai

n in

dB

TF

Pha

se in

Deg

rees

Fig. 6-4. Olinda transducer response relative to Malin transducer.

These differences in bandwidth and transient performance are not always evident, ormajor handicaps, in other forms of analysis. Figure 6-5 shows that all three transducersignals produce useful and consistent spectral characterizations for important WSCCswing modes up to perhaps 1.4 Hz. Amplitude differences for the spectral peaks can becorrected through knowledge of the transducer filtering and channel response, as can

6-6

some of the phase and timing differences. Such corrections add considerably tocomputational demands and staff workload, however, and they are rarely possible in anon-line environment. It’s better to avoid them through use of quality instrumentation.

0 1 2

-30

-20

-10

0

10

Frequency in Hertz

Aut

ospe

ctra

in d

B

Malin-Round Mtn1 MW swing

PG&E Malin Sum MW/2

PG&E Olinda MW

Malin-Round Mountain #1 MW : PG&E Malin Sum MW : PG&E Olinda MW

Fig. 6-5. Ambient noise autospectra for Malin area transducers

6.2 The Signal Environment for Power System TransducersA power system transducer is intended to extract information that has been impressedupon a set of fundamental-frequency (e.g., 60 Hz) carriers by a combination of amplitudemodulation and frequency modulation. In the simplest cases the transducer will have justone input. At the other extreme, a rms transducer for real or reactive power may have asmany as 3 voltage and 3 current signals as its inputs. There is no assurance that theirunderlying 3-phase carriers will be balanced, even during steady operation. For this andother reasons, determining the physical significance of system activity may necessitatedecomposition of the signals into “symmetrical components” plus accessory filteringspecific to their application.

The input signals may also contain components produced by mechanisms other thanmodulation of the fundamental frequency carriers (Figure 6-6). These include:

• modulated harmonics of 60 Hz

• extraneous carriers (not necessarily at harmonics of 60 Hz)

• modulated extraneous carriers

• additive transients.

Reference 6-11 surveys physical sources for such extraneous components.

6-7

ProcessingArtifacts

RMS

Transducer

Responseto Inputs

multiplier

other carriers(harmonics, etc)

other modulation(shafts, saturation, etc)

multiplier

fundamental carrier(60 Hz)

fundamental modulation(generator swings, controls, etc)

additive signals(LC resonances, etc)

+

Fig. 6-6. Signal environment for a power system transducer.

The most likely sources of frequency aliasing seem to be amplitude modulation, systemfrequency offsets, and digital decimation. Amplitude modulation can affect both analogand digital transducers. The governing relations are simple:

sin(x) sin(y) = 1

2[cos(x − y) − cos(x + y)] (6.1A)

sin(x) cos(y) = 1

2[sin(x − y)+ sin(x + y)] (6.1B)

The following examples show some consequences of these relations (see also AppendixD):

a) 1 Hz modulation of a 60 Hz carrier produces waveform components at 60±1 Hz (i.e.,at 59 Hz and at 61 Hz).

b) Re-modulation of the above waveform produces components at ±1 Hz and at 120±1Hz. Then the original modulation can be recovered by lowpass filtering.

c) 30 Hz modulation of a 60 Hz carrier produces waveform components at 30 Hz and at90 Hz.

d) 30 Hz modulation of a 120 Hz carrier produces waveform components at 90 Hz andat 150 Hz (overlapping case c).

e) Squaring any of the above waveforms produces terms at 0 Hz.

This provides a number of ways in which a signal might enter a transducer at onefrequency and be shifted to another. Those based upon amplitude modulation arediscussed more thoroughly in reference 6-14. Sampling effects in digital transducersconsiderably expand the possibilities for frequency aliasing.

Field observations confirm that transducers operate in a very demanding signalenvironment. On April 24, 1996, direct measurements were performed on enhancedtransducers at BPA’s Slatt substation [6-15]. Figures 6-7 through 6-9 show a sequence ofautospectra for A-phase current, as determined with a Scientific Atlanta SD390 4-channeldynamic signal analyzer. Related theory is available in [6-16–18]. In Figure 6-7 the peaks

6-8

near 28 Hz and 92 Hz are probably associated with a modulating source at 32 Hz (verylikely a generator shaft). These spectra are in close agreement with MATLAB analysis ofsignals extracted from the BEN 5000 digital fault recorder at Slatt, and voltage spectrawere similarly complex. Corresponding transducer spectra are shown in a later section.

Figure 6-10 extends the observations made at Slatt substation. The spectrum in this figurewas obtained at Big Eddy substation, which is directly connected to the Celilo converterof the Pacific HVDC Intertie (PDCI). The PDCI was deliberately operated in a highharmonic configuration in order to test transducer performance [6-19]. The numerousspectral peaks, many not at integer harmonics of the 60 Hz power frequency, furtherindicate just how harsh the transducer operating environment can be.

-120

-100

-80

-60

-40

-20

0

0 25 50 75 100

Frequency in Hertz

Am

pli

tud

e in

dB

A-phase current

Slatt substation, 04/24/96

100 Hz processing

28.0

0 H

z

91.9

4 H

z

Fig. 6-7. Autospectrum for A-phase current. Slatt Substation, 04/24/96.

-120

-100

-80

-60

-40

-20

0

0 100 200 400 500300

Frequency in Hertz

Am

pli

tud

e in

dB

A-phase current

Slatt substation, 04/24/96

500 Hz processing

Fig. 6-8. Autospectrum for A-phase current. Slatt Substation, 04/24/96.

6-9

-120

-100

-80

-60

-40

-20

0

0 10 20 30

Frequency in Multiples of 60 Hertz

A-phase current

Slatt substation, 04/24/96

Am

pli

tud

e in

dB

2000 Hz processing

Fig. 6-9. Autospectrum for A-phase current. Slatt Substation, 04/24/96.

Frequency in Hertz

Fig. 6-10. Autospectrum for A-phase current, HVDC controls in high harmonicconfiguration. Big Eddy Substation, 10/25/96.

Figures 6-11 and 6-12, showing frequency records for individual islands formed duringWSCC breakups in 1994, indicate that protracted operation at anomalous frequencies isanother challenge to transducer performance. In the case of Figure 6-11 the islandfrequency remains below 59.9 Hz for roughly 25 minutes. It’s possible that transducersnot designed for such operation would experience filtering or timing problems under suchconditions, and produce spurious outputs.

6-10

TIME IN MINUTES

FR

EQ

UE

NC

Y I

N H

Z

59.221

60.274

60.108

60.5

59.0

59.5

60.0

0 10 3020

Dittmer PPSM #1

Fig. 6-11. BPA system frequency following Los Angeles earthquake of January 17, 1994(BPA control center, Vancouver, Washington).

0 10 20 30 40 50

59.6

59.7

59.8

59.9

60.0

60.1

60.2

60.3

TIME IN MINUTES

FR

EQ

UE

NC

Y I

N H

ER

TZ

Kyrene substation, Phoenix AZ

Fig. 6-12. Phoenix, Arizona area system frequency following WSCC breakup ofDecember 14, 1994.

6.3 Signal Processing in Power System TransducersLet v(t) and i(t) denote the instantaneous voltage and current signals that are processedwithin a particular transducer. We will consider a transducer to be of analog type if all ofits output signals are analog, but digital if some or all of its outputs are in the form ofmulti-level digital data.

Transducers can be categorized, a bit cavalierly, into:

• “Algebraic” or “point on wave” transducers that perform simple calculations uponv(t) and i(t).

6-11

• Phasor transducers that project v(t) and i(t) onto reference waveforms, therebygenerating associated voltage and current phasors that are used in all furthercalculations (see Figure 6-13).

Sine Reference

Cosine Reference

θ

V=|V|∠θ (polar for m) = Vr + jV i (rectrangular form)

Vr

Vi

Fig. 6-13. Phasor determination via projections.

The associated logic can be organized in many different ways, and specific hardwareproducts may well contain some for each category. Whereas algebraic transducers maybe either analog or digital, contemporary phasor transducers appear to be entirely digital.

The processing in Figure 6-14 is representative of modern analog transducers that BPAuses to measure real and reactive power. The voltage input width modulates a train ofsquare pulses, which is then amplitude modulated by the current signal. The two kinds ofmodulation, PWM followed by AM, constitute an analog multiplier circuit and yield asignal that is pulse-ratio modulated (PRM). Filtering requirements are greatly reduced byfirst combining the PRM signals for all three phases.

INPUT

ISOLATION

AMPLITUDE

MODULATORIinFILTER

OUTPUT

ISOLATION Prms

INPUT

ISOLATION

COMPARATOR

CIRCUITVin

TRIANGULARWAVE

GENERATOR

Fig. 6-14. General architecture for a pulse ratio modulation (PRM) megawatt transducer.

6-12

.

GUARD

FILTERS

*RMS

CALCULATIONS

OUTPUT

DECIMATION

rms outputs

PRE-

DECIMATION

FILTERS #2

PRE-

CONVERSION

FILTERS

A/D

CONVERTER

PROTECTIVE

INTERFACE

POST-

CONVERSION

FILTERS

INPUT

DECIMATION

volt

age

sign

als

curr

ent

sign

als

(*Grey shading indicates optional element)

Fig. 6-15. General architecture for an algebraic digital transducer.

Figure 6-15 represents the functional organization (showing just one phase) for analgebraic digital transducer. This is extended in Figure 6-16, which is broadlyrepresentative of digital transducers based upon phasor calculations. The structureprovides several points where bus frequency can be estimated, and it permits use of thisestimate to adjust the reference signals upon which voltage and current signals areprojected.

SECONDARY

CALCULATIONS

REFERENCE

SIGNALS

GUARD

FILTERS

FOURIER

FILTERS

BUS FREQUENCY ESTIMATOR

SYMMETRICAL

COMPONENTS

LOGIC

multiplier

*

phasors

frequency

(*Grey shading indicates optional element)

OUTPUT

DECIMATION

PRE-

DECIMATION

FILTERS #2

PRE-

CONVERSION

FILTERS

A/D

CONVERTER

PROTECTIVE

INTERFACE

POST-

CONVERSION

FILTERS

INPUT

DECIMATION

volt

age

sign

als

curr

ent

sign

als

Fig. 6-16. General architecture for a phasor transducer.

Among the other optional features in Figure 6-16, the guard filters warrant specialmention. While their function might actually be absorbed into the post-conversion or the

6-13

Fourier filtering, the appropriate settings may change with the application. This isparticularly likely in high performance stability control, where both the control law andthe monitoring equipment should be well protected against spurious activity.

The signal processing in a phasor transducer is directly based upon Fourier analysis, andmuch the same as that used in a dynamic signal analyzer. See also Appendix E.

6.4 Criteria and Procedures for Evaluating Transducer PerformanceDistinctions are made here between the following kinds of transducer performance:

• metering performance. Emphasis upon precise measurements under normal systemconditions, network condition monitoring.

• Small-signal dynamic performance. Emphasis upon feedback control and interactionsmonitoring.

• Large-signal dynamic performance. Emphasis upon remedial action (feedforward)control, disturbance monitoring.

Technical performance factors in control applications are resolution, bandwidth, delay,accuracy, noise, protection from aliasing, other filtering considerations, and transientbehavior. The proposed test procedures will focus upon small-signal dynamicperformance, which is critical in those applications having the greatest exposure toparasitic interactions.

Distinctions are also made between kinds of information to be obtained from transducersin a control environment. As illustrated in Figure 6-17, the output of a transducer that isused to track large signal dynamics might also be low-pass filtered to display slow trendsand high-pass filtered to display small-signal activity. This assumes, of course, that signalprocessing within the source transducer is fast enough to track large signal dynamics inthe first place.

Transducer

POWER

SYSTEM

measured response y(t)

LowPass

Filter

HighPass

Filter

large-signal

activity

small-signal

activity

slow trendsSetpoint

Controls

Damping

Controls

Emergency

Controls

Fig. 6.17. Allocation of transducer signals in power system control.

6-14

The criteria for evaluating transducers in control applications are necessarily differentfrom those used in static or slowly changing measurements. The following arerecommended as high priority performance targets in control applications:

a) Bandwidth: in the range of 10–25 Hz, the higher the better.

b) Resolution and dynamic range: equivalent to 14–16 bits.

c) Delay: must be essentially constant, not to exceed 30–45 degrees within the primarycontrol band.

d) Carrier filtering (including harmonics): carrier effects will likely be visible tosophisticated analysis, but must be outside the nominal bandwidth of the transducerand small enough to remove through accessory filtering.

e) Harmonic modulation: information imposed upon power frequency harmonics abovethe first must be outside the nominal bandwidth of the transducer, and small enoughto remove through accessory filtering.

f) Positive sequence response: transducers intended to respond just to positive sequenceactivity should perform accordingly.

g) Off-frequency performance: the above criteria (a–f) should be met during sustainedramps and offsets of system frequency.

h) Unbalanced operation: the above criteria (a–f) should be met during sustainedunbalances of three phase voltage and/or current.

The following are recommended as lower priority performance targets for controlapplications:

i) Accuracy: something on the order of 0.5% of reading is usually sufficient.

j) Fixed offset: usually measured, and often removed by highpass filtering.

k) Drift: often removed by highpass filtering. If drift is substantial, it must not vary sorapidly as to mimic power system activity.

l) Instrument noise: must not have strong peaks within the primary control band, andmust be small enough to remove through accessory filtering.

It should be noted that no criteria are recommended for discriminating between additivesignals on the power system (such as network resonances) and signals associated withcarrier modulation. Desirable as such a capability would be, it’s unlikely that existingtransducers can provide it.

Appendix F describes laboratory evaluation of transducers and Appendix G describesfield evaluation of transducers.

6.5 Transducer Modeling and SimulationLaboratory tests and field examination of transducer performance have been reinforcedthrough the use of computer models. The general approach involves:

6-15

• Use of SPICE computer software [6-21,6-22] to examine interface and circuitperformance.

• Use of MATLAB computer software [6-23] to examine the generic signal processing

Models were developed for the (analog) PRM megawatt transducer of Figure 6-14, plusdigital megawatt transducers of both algebraic and phasor types (Figures 6-15 and 6-16).The MATLAB codes permit direct changes to signal processing parameters, such as filtertype and settings, and they support a broad menu of test waveforms.

Appendix H provides results of transducer modeling and simulation, and related analysis.

6.6 Digital Transducers and Phasor MeasurementsThe chapter introduction listed desired characteristics for next-generation transducers.This section assesses available products having potential for advanced stability control.

As the term is used here, the distinguishing characteristic of phasor technology andphasor transducers is the explicit calculation of the phasors themselves. Apart from thespecial values that phasors themselves may have in stability control, it’s clear that anytechnology capable of calculating them is a good candidate for developing bettertransducers. It’s equally true that any well-filtered algebraic digital transducer canprobably be converted into a phasor transducer.

It’s also apparent that that the desired class of transducers represents a functionalextension of the conventional technology, not just an improvement. A transducer that isdirectly networkable, and that performs measurements in synchronization with someprecise global reference, can be neither developed nor evaluated without considering itsrole in the overall measurement network. Also, by using the global reference in thephasor projection, all phasors in the network provide consistent angle information. Theessential integrity of phasor processing, however, is valuable even when global phaseangles are not produced.

The WAMS project [6-10] assessed the several digital transducers.

Macrodyne Phasor Measurement Unit (PMU) [6-24–26,6-37]. Considered as anindividual device, this is a phasor transducer plus:

• Optional synchronization against precise time references.

• An evolving interface for direct local networking.

• Local recording capabilities as a basic “snapshot” monitor.

General performance features of the PMU include:

• An input sample rate of 12 samples per cycle (720 samples per second at 60 Hz) afterprefiltering, with 16 bit digitization. All sampling is referenced against nominalsystem frequency, not actual.

• Output sample rates of 30, 12, and 6 samples per second, user-selectable.

6-16

• Voltage and current positive sequence phasors and bus frequency as standard outputs;quantities such as rms power or apparent resistance must be calculated later.

The PMU is expressly designed to operate in wide-area networks. Operational details andobserved performance are described in [6-13,6-25,6-27,6-37].

7700 ION Programmable Transducer System, produced by Power Measurement Ltd.(PML) [6-28]. The 7700 ION (Integrated Object Network) functions as a high bandwidthalgebraic digital transducer and, at a lower output rate, as an FFT-based harmonicanalyzer.

General performance features of the 7700 ION include:

• Input sample rate of 128 samples per cycle (7680 samples per second) afterprefiltering, usually with 12 bit digitization. All sampling is referenced against a fastrunning estimate of system frequency.

• Output sample rates ranging to 60 samples per second, according to type and userselection.

• A very wide range of rms outputs, programmable by the user. Present logic providesvoltage and current phasors at low rates only, via the harmonic analysis and using alocal reference.

The ION 7700 is highly evolved for operation in local area networks, which includecentral recording plus modem connections into wide area networks.

The Dynamic System Monitor (DSM), produced by Power Technologies Inc. (PTI) [6-29]. In this case phasor transducer logic is imbedded into a general purpose monitor.General characteristics of the DSM’s transducer include:

• An input sample rate of 4 samples per cycle (nominally 240 samples per second) afterprefiltering, with 16 bit digitization. All sampling is referenced against a runningestimate of system frequency.

• An output sample rate of 1 sample per cycle, which the user can decimate underprogram control.

• A very wide range of rms outputs, programmable by the user. Voltage and currentphasors can be obtained for the same global references that are accessed by the PMU.

The DSM is primarily designed to operate independently, but with modem connectionsinto wide area networks.

While new digital transducer technology is appearing on the market with increasingfrequency, the three devices above are well established in the field and thereby of specialinterest. They are also different enough in their processing details to span a good range ofthe basic possibilities. Evaluating these devices—or, more precisely, the technologyapproaches that they represent—remains an ongoing process.

6-17

6.7 The Transducer as an Intelligent Electronic DeviceWe have shown that transducers and transducer logic take many different forms, and arecombined into products with different functionality combinations. In short, the term“transducer” no longer has very explicit meaning when the base technology is digital.The Macrodyne PMU is an outstanding case of this. It’s a transducer for producing rmssignals, a simple monitor, and a building block for wide-area measurement networks. Atanother extreme, it is not unusual for a modern excitation controller to contain transducerlogic within a digital control law. Similar logic, sometimes in optical form, is alsoappearing in such mundane devices as electrical bushings and circuit breakers.

It can be misleading to call several different things “transducers” when the functionalitythey offer are so diverse. Similar problems are encountered with power system monitors,controllers, and even the sensors that provide signals to higher levels of the measurementsystem. There is a useful trend now to just designate any such device as an “intelligentelectronic device,” or IED [6-28,6-30].

From this perspective a wide-area measurements network is an integrated structure ofIEDs, with sensing and transducing logic occupying the lower hierarchies. SelectingIEDs with the right functionality combinations lies at the heart of the value engineeringprocess.

6.8 Role of Communication Channels in Wide-Area ControlA fully evolved stability controller for wide area dynamics requires access to signals ofthe following kinds:

• modulation signals used to directly shape the controller output u(t).

• system status flags used in

- rule-based control laws (e.g., parameter scheduling)

- coordination with other controls

- remote controller supervision

• secondary response signals for

- direct testing of power system dynamics and controller effects

- local monitoring of controller performance

- use as alternate or supplemental modulation signals

Signals are also required from the controller to operation centers, and perhaps to otherlocations, where its status and performance are supervised and coordinated with those ofother controls.

The signals used as modulating inputs are the most demanding in terms of quality,reliability, and security. These needs are most easily met if the signals are producedlocally to the controller site. Requiring this in advance, however, can make controllersiting a difficult robustness issue. There are many aspects of the controller environment

6-18

which cannot be predicted from model studies, and which may not be measurable untilthe controller itself is available for system dynamics testing.

Channels for modulation signals can, superficially, be categorized as analog or digital.For this discussion, an analog channel is one that accepts an analog signal as an input andcarries the signal in large part as a continuously varying analog signal. Analog channelsusually offer the advantages of high bandwidth relative to that of the measured signal,minimal communication delay, and reasonable immunity to undetected tampering. Theyalso tend to be noisy, and maintenance intensive. Digital channels, by contrast, involve aconversion of the signal to digital format and a commensurate increase in delay. Theyalso have a lower signal bandwidth for a given channel bandwidth but require lesschannel calibration. The digital format also allows noise free data recovery and positiveverification of data integrity.

While digital channels do not experience “noise” in the same sense as analog channels,they have a counterpart in occasional message loss. This calls for some kind of datarepair, analogous to noise filtering in analog technology. At present, digitalcommunications may also be more expensive than analog for the same bandwidth.Modems, if present, introduce communication delays and expose the information systemto penetration by unauthorized persons.

At the very lowest level, all communication systems are analog in the sense that thephysical processes can assume an infinite number of states. At higher levels, digitalcommunications modulate analog processes between a finite number of states (often justtwo states) that the detection logic is designed to recognize. Distortion and noise at theanalog level can produce errors in demodulation and, thereby, in communication ofdigital data.

Traditional analog communications take an analog signal from a transducer andtransports it as a continuously present and continuously varying voltage, current, phaseshift, or frequency shift. There are no delays other than those produced by distance andby filter effects. There are few artifacts in the information equivalent to the aliasing andquantizing errors sometimes introduced in digital systems. Means for separating noisecomponents in the signal from the information are less effective, though, and it is moredifficult to detect dropouts. Channel gains and offsets directly enter the received signals,so analog channels require frequent and precise calibration to maintain accuracy.

Modern communications frequently convert analog signals to digital quantities for longdistance transmission. The transport system can be as simple as a pair of wires or ascomplex as multi stage exchange involving satellite, microwave, and fiber optic links.This is largely transparent to the user. But, while this hybrid approach mitigates some ofthe difficulties found in completely analog systems, it can also cause new problemsassociated with digital elements of the overall system and with digital/analog interfaces.

Telephone systems are a case in point. Once entirely analog, they have beenprogressively converted to digital technology, first at the backbone (long distance) leveland more recently at the local level. This mixture of technologies means that datatransmission over telecommunication systems may encounter one or more conversionsbetween analog and digital formats. The earliest modems transmitted digital signals on

6-19

analog links by shifting the phase or frequency of a tone that was detected as digital 0’sand 1’s at the other end. There was no added communication delay other than the timeneeded to assemble a set of binary digits into a word for processing. More sophisticatedcoding has now been developed to make better use of the (analog) channel capacity. Theresult is more digital capacity, but at the expense of increased processing delay. Tomaintain the high data rate modems must “train” with each other to reduce errors. Indoing this they monitor communication errors and re-train if the errors increase unduly.This can cause an unanticipated break in communication service. Breaks can also occurthrough data re-transmission commanded by error detection logic.

Problems aside, the issue is not digital technology versus analog. Rather, the issue is howto plan and manage the transition to digital technology. Appendix I describes utilityexperience with older analog communication channels. The following section describesutility experience, based upon observed performance of a phasor measurement systemspanning a broad region of the western North American power system.

6.9 Observed Performance of Digital Communications in the BPA PhasorMeasurement Network

This section shows the performance of digital communication channels within BPA’sphasor measurement network. All channels are frequency division multiplexedmicrowave, owned and operated by BPA.

The performance data were obtained from test insertions of the Chief Joseph dynamicbrake on September 4, 1997 [6-27]. Five Phasor Measurement Units (PMUs)communicated data to a Phasor Data Concentrator (PDC) located near the DittmerControl Center in Vancouver, Washington, USA (see Figure 6-18). The PDC wasdeveloped by Ken Martin of BPA.

The PMU locations were Grand Coulee, John Day, Malin, Colstrip, and Sylmar.

The PMU at Sylmar belongs to the Los Angeles Department of Water and Power. EachPMU was configured to produce one positive sequence voltage phasor and four positivesequence current phasors at a rate of 30 samples per second (sps).

The PDC data acquired for the test consists of 23 raw data files (45 megabytes) spanningtwo recording intervals of roughly 80 minutes each. Most of these files containoccasional “outliers” in the data. These usually represent data packets (messages) thatwere lost in the digital communication system, or some brief loss of synchronism atmodem level. Accessory data from the PDC indicate these defective data precisely.

Figure 6-19 indicates that these outliers often tend to be conspicuous in the signalsthemselves, as points very near zero. The signal, which extends across all of recordinginterval #1, shows just 8 outliers among 144,000 rms power calculations. It’s possiblethat some of these represent defects in only the voltage or the current phasor, rather thanboth.

6-20

jfh

MEXICO

CANADA

GPS Synchronization& Timing

DITTMERCONTROLCENTER

MALIN

SYLMAR

PMU

COLSTRIP

PMU

JOHN DAY

PMU

GRANDCOULEE

PMU

PDC #1

PMU

HVDC Terminal

Fig. 6-18. Configuration of BPA’s Phasor Measurement Network for brake test ofSeptember 4, 1997.

For analysis, outliers near zero are easily patched through linear interpolation. The signalin Figure 6-20 demonstrates that so elementary an approach is not always enough,however. In this case the repair algorithm recognized a “blank” segment of 498 pointsand performed a linear interpolation across it. It also recognized and repaired 5 laterpoints (near 4000 seconds), but it is not equipped to deal with the rather suspicious datathat lies between the two outlier segments that it did recognize. The source file for thisrecord shows similar defects in all data extracted from the Colstrip PMU within thisparticular time frame.

Considerably more can be done to detect, flag, and (where possible) repair bad data at thesignal analysis level. However, most of this is better done at PDC level, or on the basis ofdata validity tags produced by the PDC as accessory output. Effective standards andmechanisms for this are required at both levels.

6-21

0 50K 100K 150K0

200

400

600

800

1000

1200

1400Raw Plot for Malin Round Mountain #1 MW

Time in Samples

Brake insertion #1

1 blank point

1 blank point

6 blankpoints

Brake insertion #1, 09/04/97Data collected on Dittmer PDCsample rate = 30/second

Fig. 6-19. Raw data from PDC recording segment #1.

0 2000 4000 6000 80000

2000

4000

6000

Time in Seconds

Colstrip Broadview #1raw PDC file = 09050425.MAT

patched data

5 outliers

blank data(498 points)

suspectdata

Pha

sor M

ag

nitu

de

linearinterpolation

Fig. 6-20. Partial repair of a PDC file with modem retraining.

For control, detection of bad data typically causes freezing of the signal at the last goodvalue. Sustained bad data may cause wide-area control suspension. Provisions for(degraded) local control for communication failure is normally required.

6-22

6.10 Future Digital Communication for Stability ControlPower companies and telecommunication companies are rapidly installing long distancefiber-optic communication. This is a very important development that greatly facilitateswide-area stability control feasibility. Some designs are “self-healing,” with transfer to analternate path in 50–120 ms for a failure.

BPA has several phasor measurement links employing modems and analog microwave,and one link employing fiber optics. The measured modem/analog time delay (latency)averaged about 70 ms. The measured fiber optic link time delay was about 21 ms,corresponding to about 7° of a 1 Hz signal.

Other emerging communication techniques promise fiber optic like performance. One islow earth orbit satellites [6-34] and possibly other wireless technology. Another is digitalcommunications over power lines. These techniques may make direct load control (e.g.,heaters, air conditioners) for stability more practical. Direct load control is described inthe next chapter.

6.11 Optical SensorsThe transducers described above normally use the outputs of conventional instrumenttransformers (magnetic current and voltage transformers, and capacitor voltagetransformers). Optical voltage and current sensors, however, are commercially availablefrom several manufacturers. Voltage and current sensors may be combined in a singledevice. Field evaluations have been successful [6-35,6-36].

Optical sensors may be used with digital IEDs in the substations of the future.

The advantages of optical sensors are only indirectly related to use of digital transducersfor stability controls. Advantages include smaller device size and weight, elimination ofhazardous oil-filled transformers, electrical isolation, elimination of substation secondaryelectrical cabling, elimination of instrument transformer burden limitations, widedynamic range, wide bandwidth, high accuracy, potentially lower cost, and compatibilitywith digital technology.

Challenges are mainly economic, related to change out of existing instrumenttransformers, compatibility with legacy electromechanical relays and meters, and thevolume production needed for cost reduction.

Acknowledgement: Much of the material in this chapter is based on findings of theDOE/EPRI Wide Area Measurement Systems (WAMS) project [6-9,6-10]. Material isreproduced here with the permission of BPA.

References6-1 J. F. Hauer, “Robust Damping Controls for Large Power Systems,” IEEE Control

Systems Magazine, pp. 12–19, January 1989.

6-2 CIGRE Task Force 38.01.07, Analysis and Control of Power System Oscillations.CIGRE Brochure 111, December 1996.

6-23

6-3 J. F. Hauer, “BPA Experience in the Measurement of Power System Dynamics,”Inter-Area Oscillations in Power Systems, IEEE Publication 95 TP 101, pp. 158–163, 1995.

6-4 J. F. Hauer and J. R. Hunt, in association with the WSCC System OscillationsWork Groups, “Extending the Realism of Planning Models for the Western NorthAmerica Power System,” V Symposium of Specialists in Electric Operational andExpansion Planning (V SEPOPE), Recife (PE) Brazil, May 19–24, 1996.

6-5 J. F. Hauer, W. A. Mittelstadt, R. J. Piwko, B. L. Damsky, and J. D. Eden“Modulation and SSR Tests Performed on the BPA 500 kV Thyristor ControlledSeries Capacitor Unit at Slatt Substation,” IEEE Transactions on Power Systems,Vol. 11, pp. 801–806, May 1996.

6-6 C. W. Taylor and D. C. Erickson, “Recording and Analyzing the July 2 CascadingOutage,” IEEE Computer Applications in Power, Vol. 10, No. 1, pp. 26–30,January 1997.

6-7 J. F. Hauer, D. J. Trudnowski, G. J. Rogers, W. A. Mittelstadt, W. H.Litzenberger, and J. M. Johnson, “Keeping an Eye on Power System Dynamics,”IEEE Computer Applications in Power, pp. 50–54, October 1997.

6-8 D. N. Kosterev, C. W. Taylor, and W. A. Mittelstadt, “Model Validation for theAugust 10, 1996 WSCC System Outage,” IEEE/PES paper PE-226-PWRS-0-16-1997, to be published in IEEE Transactions on Power Systems.

6-9 W. A. Mittelstadt, P. E. Krause, P. N. Overholt, D. J. Sobajic, J. F. Hauer, R. E.Wilson, and D. T. Rizy , “The DOE Wide Area Measurement System (WAMS)Project—Demonstration of Dynamic Information Technology for the FuturePower System,” EPRI Conference on the Future of Power Delivery, Washington,D.C., April 9–11, 1996.

6-10 J. F. Hauer, W. A. Mittelstadt, W. H. Litzenberger, C. Clemans, D. Hamai, and P.Overholt, Wide Area Measurements for Real-Time Control and Operation ofLarge Electric Power Systems: Evaluation and Demonstration of Technology forthe New Power System. Report prepared for U.S. Department of Energy byBonneville Power Administration and Western Area Power Administration, April1999. This report and attachments are available from BPA on compact disk.

6-11 M. R. Irvani, et al., “Modeling and Analysis Guidelines for Slow Transients: Part1 (Torsional Oscillations; Transient Torques, Turbine Blade Vibrations; Fast BusTransfer),” IEEE Transactions on Power Delivery, Vol. 10, No. 4, pp. 1950–1955, October 1995.

6-12 B. J. Hickman and J. F. Hauer, “General Characteristics of Power SystemTransducers,” WAMS Working Note, December 20, 1995, attachment toreference 6-10.

6-13 J. F. Hauer, “Validation of Phasor Calculation in the Macrodyne PMU forCalifornia-Oregon Transmission Project Tests of March 1993,” IEEETransactions on Power Delivery, Vol. 11, No. 3, pp. 1224–1231, July 1996.

6-24

6-14 J. F. Hauer, “Signal Processing Aspects of Power System Transducers,” WAMSWorking Note, June 18, 1996, attachment to reference 6-10.

6-15 J. F. Hauer, “A Preliminary Report on Transducer Measurements Performed atSlatt Substation on April 24, 1996,” WAMS Working Note, June 5, 1996,attachment to reference 6-10

6-16 E. O. Brigham, The Fast Fourier Transform and Its Applications, EnglewoodCliffs, NJ: Prentice-Hall, 1988.

6-17 J. S. Bendat and A. G. Piersol, Engineering Applications of Correlation andSpectral Analysis, New York: John Wiley, 1980.

6-18 J. G. Proakis and D. G. Manolakis, Digital Signal Processing—Principles,Algorithms, and Applications (Second edition), New York: McMillan, 1992.

6-19 M. K. Donnelly, R. Bunch, J. Dagle, and B. Hickman, “Performance of the PML7700 ION Programmable Transducer System, as Tested at BPA's Big EddySubstation on October 25 1996,” WAMS Working Note, March 12, 1997,attachment to reference 6-10

6-20 J. F. Hauer, “Nonintrusive Procedures for Measuring Dynamic Performance ofEnhanced Transducers at Slatt Substation,” WAMS Working Note, December 7,1995, attachment to reference 6-10

6-21 Thomas W. Thorpe, Computerized Circuit Analysis with SPICE, John Wiley &Sons, 1992.

6-22 P. W. Tuinenga, SPICE, A Guide to Circuit Simulation & Analysis Using PSpice.Third Edition, Prentice Hall, 1995.

6-23 MATLAB – High-Performance Numeric Computation and Visualization Software(Reference Guide), The Math Works, Inc., Natick, Mass., 1992.

6-24 A. G. Phadke, “Synchronized Phasor Measurements in Power Systems,” IEEEComputer Applications on Power Systems, pp. 10–15, April 1993.

6-25 R. E. Wilson, P. S. Sterlina, and B. W. Griess, “GPS Synchronized Power SystemPhase Angle Measurements Recorded During 500 kV Staged Fault Testing,”Third Virginia Tech Conference on Computers in Electric Power Engineering,Arlington, VA, October 27–29, 1993.

6-26 K. E. Martin, “Phasor Measurements on the BPA Transmission System,” BPAWorking Note, May 1997.

6-27 J. F. Hauer et al., “Research Database from BPA’s Phasor Measurement Networkfor Test Insertions of the Chief Joseph Dynamic Brake on September 4, 1997,”WAMS Information Manager Working Note, March 3, 1998.

6-28 R. Carolsfeld, “To Measure is to Control,” Electrical Business, pp. 14–15,January 1997.

6-25

6-29 H. K. Clark, R. K. Gupta, C. Loutan, and D. R. Sutphin, “Experience withDynamic System Monitors to Enhance System Stability Analysis,” IEEETransactions on Power Systems, Vol. PWRS-7, pp. 693–701, May 1992.

6-30 H. L. Smith, “Substation Automation Problems and Possibilities,” IEEEComputer Applications in Power, Vol. 9. No. 4, pp. 33–36, October 1996.

6-31 J. F. Hauer, et al., Research Database from BPA’s PPSM Network for TestInsertions of the Chief Joseph Dynamic Brake on September 4, 1997, WAMSInformation Manager Working Note, March 3, 1998, attachment to reference 6-10.

6-32 R. L. Cresap, D. N. Scott, W. A. Mittelstadt, and C. W. Taylor, “Damping ofPacific AC Intertie Oscillations via Modulation of the Parallel Pacific HVDCIntertie,” CIGRE 14-05, 1978.

6-33 R. L. Cresap, D. N. Scott, W. A. Mittelstadt, and C. W. Taylor, “OperatingExperience with Modulation of the Pacific HVDC Intertie,” IEEE Transactionson Power Apparatus and Systems, Vol. PAS-98, pp. 1053–1059, July/August1978.

6-34 B. Miller, “Satellites Free the Mobile Phone,” IEEE Spectrum, Vol. 35, No. 3,March 1998 (http://www.spectrum.ieee.org/spectrum/mar98/features/leo.html).

6-35 J. Tillett, J. Pease, J. Hall, and D. Bradley, “Experience with Optical PTs and CTsfor Relaying and Metering,” Proceedings of Western Protective RelayConference, October 23–25, Spokane, Washington USA.

6-36 D. Chatrefau, “Application of Optical Sensors in Extra High VoltageSubstations,” GEC Alsthom T&D Review, pp. 17–24, 1/97.

6-37 K. Martin and R. Kwee, “Phasor Measurement Unit Performance Tests,”Proceedings of Precise Measurements in Power Systems Conference, Arlington,Virginia, November 8–10, 1995.

Chapter 7

Applications of Advanced Controls

This chapter gives examples of engineering projects where advanced control methodshave been used or studied. The first example is from Brazil where thyristor controlledseries compensation (TCSC) is used to damp interarea oscillations. Section 7.2 presentsthe control analysis of a potential 500-kV TCSC installation in China. Section 7.3explores how new distributed-measurement technology can be used to improve dynamicand transient system stability. Section 7.4 describes active-load modulation to improvestability. Section 7.5 presents energy source power system stabilizers.

7.1 Brazilian North–South Interconnection—Application of ThyristorControlled Series Compensation (TCSC) to Damp InterareaOscillation Mode

This example deals with a pioneer commercial application of TCSC to damp the lowfrequency interarea oscillation mode in the Brazilian north–south interconnection [7-1].The north–south interconnection connects Imperatriz substation (in the State ofMaranhão) to Serra da Mesa (in the State of Goiás). The interconnection is a single 500-kV line and is 1,020 km long. The line is designed to transmit up to 1,300 MW, withsuitable operation required from no load up to maximum flow in both directions. Theinterconnection was commissioned in early 1999 and reduces the risk of energy deficits.A TCSC control system for transient and dynamic stability improvement was designedand, together with extensive study results, formed the basis for the TCSC locations andequipment specification. Use of two small TCSCs (6% compensation each) proved to bevery effective in damping the interarea mode, and eliminated the technical restriction onthe AC transmission alternative.

Main aspects. There are two main electric power systems in Brazil which were notpreviously interconnected: the south/southeast (or south system) and the north/northeast(or north system) systems. They are essentially hydroelectric systems and include morethan 95% of the total national production and consumption. The installed generationcapacity in South/Southeast and North/Northeast systems is about 48 GW and 14 GW,respectively. See Figure 7-1.

Technical and economical feasibility of the interconnection was studied since 1992. The“North–South Interconnection” will exploit hydrologic diversity between the systems,achieving energetic benefits estimated at about 600 MW-year. Power flows will occur inboth directions, depending on the actual hydrologic conditions.

Two transmission alternatives were considered and analyzed to establish the North–SouthInterconnection: a DC ±400-kV bipole and a single 500-kV AC compact transmissionline (4x954 MCM bundle), 1,020 km long. In both cases, the interconnection links the500-kV substation of Imperatriz (north system) to the Serra da Mesa power plant (southsystem).

7-2

Fig. 7-1. Brazilian North–South Interconnection—geographic location.

From a purely technical viewpoint, this long, low capacity interconnection between twolarge systems having different planning and operating criteria is a textbook application forHVDC transmission technology. From a strategic and political viewpoint, however, theAC transmission alternative is highly attractive for making inexpensive hydroelectricenergy available to a rapidly growing area, and for future generation developmentslocated over a vast geographic area having enormous economic potential. Sixhydroelectric plants may be built along the same route in the next two decades, and other500-kV AC transmission links are planned to cater for this additional generation.

When comparing the technical behavior of the two alternatives, it was verified that theAC solution presented a low frequency (0.18 Hz), poorly damped interarea oscillationmode. This oscillation of wide amplitude (± 300 MW) represented a serious technicalrestriction for the AC alternative. On the other hand, this alternative presented significantadvantage in terms of costs, besides the strategic and political benefits mentioned above.

Traditionally, the problem of electromechanical oscillations in the range of 0.5 to 2.0 Hzhas been solved by power system stabilizers (PSS) in the main synchronous generators.

7-3

For lower frequency modes (< 0.3 Hz), however, effective damping is a difficult task. Themain drawbacks of this solution for the north–south interconnection are listed below [7-1]:

1. Modified PSSs would be needed in all major power plants of the northeast system;

2. The modified PSSs, assumed to be of fixed structure and fixed parameters, would notalways ensure adequate damping for the north–south mode for the various scenariosconsidered in the study;

3. The frequency range of electromechanical oscillations to be damped by the modifiedPSSs is too wide to yield reliable operation;

4. Electromechanical oscillations within the northeast system (local modes) and betweenthe north/northeast systems (interarea mode) could have their damping reduced by theaction of the modified PSSs;

5. Practical limitations on maximum PSS gain at very low frequencies may reduce thedamping of these modified stabilizers.

To solve the sustained oscillation problem thyristor controlled series compensation(TCSC) was proposed in the interconnection (transmission line Imperatriz–Serra daMesa). This solution was much more efficient than PSS in providing damping for allpossible system scenarios and contingencies. One great advantage of this solution is thefact that the TCSCs are located in the link that introduces the interarea mode and they aretuned only for this mode, not having any effect on the other modes presented in thesystem. So if this link is disconnected, the TCSCs together with the interarea mode ceaseto exist. The stability of the two isolated systems (north and south) in this case isguaranteed by PSSs exactly as before the advent of the interconnection.

The TCSCs at each end of the intertie are modulated using local line power measurements.Figure 7-2 shows simulation results. Commissioning tests verified the powerful dampingperformance of the TCSCs [7-2].

7.2 Analysis and control of Yimin–Fengtun 500-kV TCSC systemReferences 7-3–5 present research done for the thyristor controlled series compensation(TCSC) to be situated on the main corridor of the 500-kV transmission system ofnortheast China. Power will be transferred from Yimin plant in Mongolia, with 2200 MWcapacity, to the load centers through a 500-kV parallel transmission line covering adistance of 1300 km. The paper is motivated by the real engineering project and presentson-going research for TCSC models, control algorithms, simulation software andimplementation. Because of future development of the transmission system, the controllerdesign must be systematic and robust.

The Yimin–Fengtun TCSC projects has the following distinctive features:

• It’s located on an important corridor of the main grid. The theories and the schemesmust be applied to the real engineering project and must be easy to manipulate.

• It’s required to increase dynamic, transient and voltage stability.

7-4

600

800

1000

1200

0 5 10 15

Pow

er -

MW

Time - seconds

Fig. 7-2. Simulation of fault with line outage in south system [7-1]. Thin line withoutTCSCs, thick line with TCSCs.

• The controller must be robust. It should adapt to not only the variation of operationconditions, but also future changes of grid topology. It’s important to have controlschemes that require local signals only and are independent of system models.

• Because of the project’s importance, the research should be broad.

Without a TCSC, the system suffers severe transient and dynamic instabilities. The studyshows that the TCSC with proper control schemes can schedule power flow flexibly, andimprove transient and dynamic stability. The influence of the voltage protection of themetal oxide varistor (MOV) is included in the control model and design. Auto-disturbance rejection control (ADRC), fuzzy control, and nonlinear adaptive scheme arestudied. Simulations show the effectiveness of the control schemes. The combined effectof electromagnetic and electromechanical transients has been studied.

7.3 Wide-Area Stability ControlNew distributed measurement technology using the global positioning system andaccurate phasor measurements units have developed steadily in recent years to becomethe most powerful source of wide-area dynamic information. Reference 7-6 explores newways of putting this extended real-time knowledge of the power system behavior into useby means of supplementary feedback loops which improve dynamic and transient systemstability and, ultimately, increases the transmission capacity.

The design of such advanced controllers is based on a two-stage methodology. The firststep is built on a powerful pulse response-based, numerical sub-space, state-spaceidentification algorithm to identify a reduced-order small-signal MIMO model of the

7-5

open-loop system. The second step is to select an appropriate control structure, and thentune the stabilizer parameters accordingly. To tackle the most difficult situations, thearchitecture selected comprises several dynamic feedback loops, each consisting of ahigh-order differential filter. Controller tuning is then performed by minimizing aselective modal performance index in the parameter space.

Adding stability and robustness constraints greatly improves the engineering significanceof the resulting design. For illustration, a three-loop stabilizer was designed for a majorsynchronous-condenser station in an actual power system that simultaneously uses twoglobal and one local input signals. Both linear and nonlinear simulation results clearlydemonstrate the added value of wide-area information when properly included in powersystem stabilizer design.

External grid(25,000 MW)

750 MVA SC

Duvernay

Ref. Area

PSS

SCADAΣ

Vref

Wide-AreaMeasurements

θ8-θ ref

θ4 -θ7

f294

+

+

Fig. 7-3. Decentralized/hierarchical PSS located at the Duvernay synchronous condensers(SC).

The architecture of the Hydro Quebec test system used in reference 7-7 is recalled inFigure 7-3 .The target PSS is at the Duvernay synchronous condenser in the referencearea. This site was chosen because it shows the highest controllability index over thebroadest frequency range. The three inputs of the PSS are the following:

[ ]Reffy θθθθ −−= 874294

where 294f is the local bus frequency at Duvernay, and 74 θθ − and Refθθ −8 are angle

shifts between the subscript areas. Based on three typical contingencies, Figures 7-4 and7-5 provide some interesting clues as to what added value should be ascribed to theinformation exchange paths outlined on Fig. 7-3. On the first contingency, the local loopalone was sufficient to stabilize the system. However, it was unable to do the same for thesecond and third, although its positive action provided 5–10 seconds relief before actualbreakdown. Therefore, information exchange really has some monetary value, which insome cases could pay for the implementation costs and cover the additional risks inherentin long-distance telemetry.

7-6

0 5 10 15 20 25 30 35

−700

−600

−500

−400

−300

−200

−100

0

s

Duve

rney

freq

uenc

y dev

iation

(mHz

) Unstable fault in the western corridor (BUS #783)

0 5 10 15 20 25−160

−140

−120

−100

−80

s

Angle

shift:

8−R

ef (D

eg)

No PSS LPSS+GPSSLPSS

Fig. 7-4. First contingency: The local loop alone prevents the system from collapsing.

0 5 10 15 20 25−160

−140

−120

−100

−80

−60

s

Angl

e Sh

ift o

f are

a #8

w.r.

t Ref

.(Deg

.) (a) Unstable fault in the eastern area (Bus #706)

0 5 10 15 20 25−160

−140

−120

−100

−80

−60

s

Angl

e Sh

ift o

f are

a #8

w.r.

t Ref

.(Deg

.) (b) Unstable fault in the western area (Bus #780)

No PSS LPSS+GPSSLPSS

Fig. 7-5. Second and third contingencies: The local loop alone substantially improves thesystem performance, but doesn’t prevent instability.

7-7

7.4 Active-Load Modulation for Stability ControlThe first subsection describes large-scale load modulation and the second subsectionpresents field tests at a small hydro station in Sweden.

Large-scale load modulation. Reference 7-7 describes how angle stability can beimproved by large-scale active-load modulation. Analysis, operating experience, andsimulation of a large power system is used to demonstrate that active-load modulation canimprove system dynamic performance to a large extent, with just a fraction of the baseload available for control. At a time when the cost effectiveness of power electronicdevices for damping interarea oscillations is constantly being questioned, it’s natural tolook to active-load modulation as a potential alternative method of ensuring gridreliability. In developing the case, it was found that continuously modulating loadstabilizers need global signals for full effectiveness. Although more difficult to design,implementation of discontinuous control schemes show good prospects, especially fordecentralization and robustness against communication delays.

Damping of Power Oscillations by Load Switching—Field Tests at Hemsjö HydroPower Station. Reference 7-8 presents field tests performed at the hydro power stationHemsjö Övre the night of 24 and 25 September 1996. The tests were done to investigateif load switching could be used to damp power oscillations. The results show that loadswitching is an excellent method of damping power oscillations.

The idea is to switch a resistive load so it counteracts power oscillations. The angledifference between the external net and the estimated generator internal EMK was used tocontrol the load switching. The load used was pure resistance without any dynamics andwas dedicated for this purpose.

To make the generator susceptible to power oscillations, the grid configuration waschanged so the hydro power station had to feed its power through a weak distributionsystem before connecting to the main grid

A way to verify the damping effect of load switching is to change the sign in thecontroller, corresponding to a 180 degrees phase change. A change of sign in a well-tunedregulator can induce power oscillations. During the first 10 seconds in Figure 7-6 and 7-7the regulator sign was changed. The figure clearly shows that the load switching builds upa power oscillation with increasing amplitude. Two measurements were done to comparedamping. The first case is without load switching and in the second case is with loadswitching. Figure 7-6 with time > 10 second shows the damping without load switching.

7-8

0 5 10 15 20 25 30

0.3

0.4

0.5

t20; Generator Power in MW

Time [s]

P_ge

n [MW

]

0 5 10 15 20 25 300

0.005

0.01

0.015

0.02

0.025t20; Power to Controlled Load in MW

Time [s]

P_loa

d [MW

]

Fig. 7-6. Excitation with load switching (1-9s) and thereafter no switching.

Figure 7-7 shows the damping when load switching is used. It is evident that controlledload switching improves damping considerably. Note that the power of the switched loadis only a fraction of the oscillation amplitude.

0 5 10 15 20 25 30

0.3

0.4

0.5

t24; Generator Power in MW

Time [s]

P_ge

n [MW

]

0 5 10 15 20 25 300

0.005

0.01

0.015

0.02

0.025t24; Power to Controlled Load in MW

Time [s]

P_loa

d [MW

]

Fig. 7-7. Excitation by load switching (1–9 s), thereafter no switching (9–14 s), thendamping by controlled load switching (14–30 s).

7.5 Active Power Modulation of Generators and Energy Storage forOscillatory Instability Control

Energy Source Power System Stabilizer (ESPSS). The objective of the ESPSS is todamp low frequency electromechanical oscillations between large interconnected powersystems. Field tests and monitoring have demonstrated the ESPSS performance in sensingsystem disturbances, and in controlling the power of batteries [7-9] and steam-turbinegenerators in response to these system oscillations.

The ESPSS can be applied on energy storage systems such as superconducting magnetic

7-9

energy storage (SMES) or battery energy storage systems. The power import capabilityand the reliability can be increased significantly by damping the interarea power systemoscillations that often limit such imports.

In the case of a battery energy storage system (BESS), ESPSS processes the frequencydeviation signal or similar signal to control the megawatt output or input of the batteries.The ESPSS, which controls the real power output to counteract these oscillations, canprovide effective damping. The ESPSS can be applied on the battery or superconductingenergy storage by controlling the power conditioning system (PCS) which converts powerbetween AC and DC. The state-of-the-art PCS using Gate-Turn-Off (GTO) thyristors arevery fast acting and have the capability to accept both MVAr and the MW power orders.The ESPSS controls the MW output only.

Tests conducted at the 10 MW Chino battery energy storage system [7-10] demonstrateddamping capability with measurable results. However, a much larger BESS or SMES isrequired to effectively damp and stabilize the system. Since the aim is to provide dampingtorques to generators, the most effective location of energy storage is close to generatorsparticipating in low frequency oscillations.

ESPSS installation on electric power generators. In damping interarea modesconventional PSS essentially modulates voltage-sensitive load, and the effectivenessdepends on the location and characteristics of load, on the tightness of voltage control,and on the mode shape [7-11]. These factors affect the component of electrical torque inphase with (modal) speed that produces damping.

For generators, the ESPSS differs from conventional excitation equipment PSS in that itacts on the mechanical input power of the generator. It can be effective and robust indamping low frequency modes present in the speed signal, with less dependence onvariable network and load characteristics, and generator loading.

The ESPSS concept is to produce damping more directly by modulating the mechanicalinput power instead of generator voltage and reactive power. This is by adding a speed orfrequency deviation based signal into the governor valve controls. Similar to PSS, theinput signal can be derived from speed/frequency and electrical power measurements.Appendix J further describes mechanical versus electrical side damping.

Field tests conducted on a turbo-generator with a state-of-the-art governor showed thatsteam-turbine governors can respond fast enough to provide damping of low frequenciesoscillations (0.2–0.8 Hz range). Thus the ESPSS concept can be extended to the othersteam-turbine governors. With a large power source it would be possible to damp theoscillations with even a small change (5 percent) of the generator output.

At Alamitos Generating Station in California, generators 5 and 6 steam turbines are cross-compound units and the steam flow is controlled on the high-pressure side. The steamcontrol is obtained from eight valves. The opening and closing of the valves arecontrolled to obtain maximum operating efficiency and control. Tests were conducted byinjecting the modulating signal in one and two different valves of these eight valves.Modulating two valves gave almost twice the modulated power output change comparedto one valve. By modulating two valves, modulation of 5 percent of the turbine power

7-10

(about 24 MW) can be achieved. The modulation input to the valve is dependent on thefrequency excursion from 60 Hz and can be adjusted by changing the gain of the ESPSS.

Figure 7-8 shows the gain and the phase relationship of the governor loop measured bychanging the input into the governor control board and monitoring the megawatt changein the machine output. The phase lag increases as the frequency of the modulation signalincreases. At 1 Hz, the phase shift between the injected input signal and the power outputincreases to about 100 degrees. This phase shift includes delays in the steam circuit suchas the steam chest. For damping control design, the transfer function between the valveinput and mechanical power is required, and this can be computed from measurements ofelectrical power and speed. The modulation control includes phase compensation of thesteam circuit lag so that the change in mechanical power is closely in phase withgenerator speed changes for oscillation frequencies of interest.

Figure 7-9 shows the response curve for the excitation system of a similar machine. Thephase shift in this case is increases much more rapidly, increasing to about 180 degrees atabout 0.7 Hz. However, the gain also drops rapidly making this control loop ineffective atthese higher frequencies.

Although efforts to implement these controls were made in the past, it had not beenfeasible because the governors were generally slow. Also, the frequencies that wereattempted were mostly local mode oscillations and were in the range of 1.0–3.0 Hz. Theadvanced state-of-the-art governors and the lower interarea oscillation frequencies havemade this modulation feasible.

Two ESPSS have been developed and installed at Alamitos generating units 5 and 6. TheESPSS acts only for large system disturbances. It cuts off the excitation system PSSsystem when it operates as shown in Figure 7-10.

Fig. 7-8. Governor frequency response with signal input into two valves.

7-11

Fig. 7-9. Excitation system frequency response for Alamitos generating unit.

Fig. 7-10. Functional block diagram for an integrated power system stabilizer.

Rapid modulation of fuel flow in combustion turbines. In so-called “industrial”combustion turbines, the turbine compressor and power turbine are mechanically coupledto the synchronous generator and thus turn at a speed that is constantly proportional tosynchronous speed. This arrangement, also called “single-shaft” combustion turbines,maintains constant air flow through the entire unit. With this constant air flow rate, theturbine power changes within milliseconds of changes in fuel flow into the combustorsallowing the rapid power generation change that may be used to improve transient andoscillatory instability. For instance, it’s possible for such a unit to shift from synchronous

7-12

operation at minimum generation, approximately 30 % of nameplate power, to fullnameplate power in less than a second by rapidly increasing fuel flow. The drawback tothis process change lies in the fact that the power turbine temperatures also changerapidly with changes in firing rate, and the erosion rate of power turbine blades isseriously increased with large, rapid changes in firing rate and temperature. Utilities haveexperimented with these concepts and with rapidly increasing firing rate duringemergencies such as loss of large blocks of generation, but we know of no in-serviceapplications.

Similar to the above descriptions of small (5%) modulation of steam turbines, modulationof gas turbines should be possible without damaging temperature excursions. For two-sided modulation (increase and decrease of power), the gas turbine generator would haveto be operated at lower efficiency, below maximum power. This might require anancillary service arrangement, with compensation for the lost power sales and lowerefficiency. The ancillary services could include system stability (damping), primary andsecondary spinning reserve, and increased reactive power production or reactive powerreserve.

References7-1 C. Gama, R. Leoni, J. B. Gribel, R. Fraga, M. J. Eiras, W. Ping, A. Ricardo, J.

Cavalcanti, and R. Tenório, “Brazilian North–South Interconnection —Application of Thyristor Controlled Series Compensation (TCSC) to Damp Inter-Area Oscillation Mode,” CIGRÉ, paper 14-101, 1998.

7-2 C. Gama, “Brazilian North-South Interconnection — Control Application andOperating Experience with a TCSC,” Proceedings of IEEE/PES 1999 SummerMeeting, pp. 1103–1108, Edmonton, 18–22 July 1999.

7-3 X. Zhou, et al., “Analysis and Control of Yimin–Fengtun 500 kV TCSC System,”Electric Power Systems Research, No. 46, pp. 157–168, 1998.

7-4 X. Zhou and J. Liang, “Overview of Control Schemes for TCSC to Enhance theStability of Power Systems,” IEE Proc.-Gener. Transm. Distrib., Vol. 146, No. 2,pp. 125–130, March 1999.

7-5 X. Zhou and J. Liang, “Nonlinear Adaptive Control of TCSC to Improve thePerformance of Power Systems,” IEE Proc.-Gener. Transm. Distrib., Vol. 146,No. 3, pp. 301–305, May 1999.

7-6 I. Kamwa, L. Gérin-Lajoie, and G. Trudel, “Multi-Loop Power System StabilizersUsing Wide-Area Synchronous Phasor Measurement,” presented at AmericanControl Conference, June 1998.

7-7 I. Kamwa, R. Grondin, D. Asber, J. P. Gingras, and G. Trudel, “Large-ScaleActive-Load Modulation for Angle Stability Improvement,” IEEE Transactionson Power Systems, Vol. 14, No. 2, pp. 582–590, May 1999.

7-8 O. Samuelsson and M. Akke, “On-Off Control of an Active Load for PowerSystem Damping - Theory and Field Test,” IEEE Transactions on Power Systems,

7-13

Vol. 14, No. 2, pp. 608-613, May 1999.

7-9 B. Bhargava and G. Dishaw, “Energy Source Power System Stabilizer Installationon the 10 MW Battery Energy Storage System at Chino Substation,” presented atIEEE Summer Meeting in Berlin, Germany, July 20-24, 1997.

7-10 L. H. Walker, “10-MW GTO Converter for Battery Peaking Service,” IEEETransactions on Industry Applications, Vol. 26, No. 1, pp. 63–72, January/February 1990.

7-11 P. Kundur, “Effective Use of Power System Stabilizers for Enhancement of PowerSystem Reliability,” Proceedings of IEEE/PES 1999 Summer Meeting, pp. 96–103, Edmonton, 18–22 July 1999.

Chapter 8

STABILITY CONTROLS WITHINDUSTRY RESTRUCTURING

Industry restructuring from a highly centralized hierarchical and possibly state-ownedsystem to a new model characterized by competition in generation with guaranteed accessto transmission has many impacts on power system stability.

Unlike the case of other industries such as communications and transportation whereoverloads result merely in telephone busy signals, or gridlock at toll plazas, powerconsumption is instantaneous. Power must be supplied the moment a switch is turned on.Inadequacies in the generation/transmission plant can result in system collapse withunacceptable consequences.

The dynamic performance of power systems, that is the ability of maintaining reliable andstable supply within tolerable limits of voltage and frequency, is a function of the jointcharacteristics of generation, transmission, control and protection, and loads.

In the traditional approach of integrated planning of bulk electric systems by a centralizedcompany or agency, the decision process on generation, transmission, distribution, andcontrol additions was well structured. The aim was an optimum allocation of investmentin the various segments so as to achieve a prescribed level of reliability at minimum cost.

In Brazil for example, the plans of generation additions were established almostindependently, with mere estimates on transmission feasibility. This was due to the highrelative cost of generation. Transmission planning then proceeded to accommodate analready established generation master plan.

The recent abundance of natural gas and the rapid progress of combustion turbine andcombined-cycle technology has drastically changed the economics of generation.Concurrently, the worldwide trend to deregulation has opened the power generationindustry to independent producers. The restructuring of the power industry requiresestablishing requirements for new generation equipment and controls, and requiresadministering the required ancillary services in the new operating environment.

So while competition will force the evolution of the most economic generation additions,there will still be some aspects of dynamic characteristics requiring cooperation dictatedby the effects on overall system performance. Stability control, including the equitableallocation of associated costs, is one such issue.

8.1 Some Examples of New Scenarios

8.1.1 The Brazilian electric systemThe predominantly hydro Brazilian system spans large geographical distances and hasmost of its generation remote from load centers. Unfavorable hydrological conditions

8-2

frequently call for high power transfers between regions, even during light loadconditions. Stability problems are therefore naturally aggravated during these conditions.

Up to now, the allocation of costs of stability controls has been decided jointly by theGCPS and GCOI, the national coordinating pools for planning and operation of theBrazilian power system, whose decisions are mandatory. The stability controls consideredinclude control of system oscillations (PSS), generation dropping, underfrequency loadshedding, dynamic voltage controls, HVDC controls, controlled islanding, and automaticswitching of shunt compensation.

In the old system structure stability problems were detected and resolved by GCOI/GCPS.The introduction of IPPs, cogeneration, and an ISO (Independent System Operator)changes this picture. Attributing responsibility for a given stability problem anddistributing the costs of candidate solutions are very complex issues with opposingopinions. There is therefore need to investigate these aspects in order to establish guide-lines, responsibilities, and associated costs for stability controls in a competitiveenvironment.

The Brazilian electric system has an installed capacity of 56,000 MW which ispredominantly hydro (95%), has 150,000 km of transmission lines of voltage levels from138-kV to 765-kV. The energy production is of the order of 309 TWh, with 97% beingfrom hydro. There are about 40 million consumers, with 32.5 million being residentialconsumers. The energy consumption per capita is 1,954 kWh/year for residentialconsumers.

Other Brazilian system characteristics are:

• Large capacity hydro power plants remote from the load centers.

• Large hydroelectric dams, having up to 5 years storage capacity for good regulation ofvariable inflows.

• Hydro units of large capacity: Itaipu (700 MW), G. Munhoz (418 MW), Itumbiara(380 MW), etc.

• Long transmission lines, sometimes presenting bottlenecks in some transmissioncorridors.

• Frequent operating conditions with heavy energy transfers, even during light load, dueto hydroelectric generation coordination for optimal water usage.

• High load growth (6% per year, during 1996/1997).

• Delays in construction of high capital investment power plants, with consequent needfor urgent generation expansion.

During unfavorable hydrological conditions, the transfer of large blocks of energybetween generating subsystems having hydrological diversity is carried out mainly duringlight load conditions. In some parts of the system, it’s common to have reversals in thepower flow of some transmission lines and transformers. In a few cases involving highlyunfavorable hydrological conditions, the system has operated with violations of the

8-3

existing criteria of transient stability. Note that in these cases the system must still meetthe criteria for small-signal stability to avoid spontaneous oscillations.

As stated before, there are two coordinating bodies for the expansion planning andoperation of the interconnected systems, which are composed of managers fromEletrobras and all the other Brazilian utilities—the GCPS (Coordinating Group of SystemPlanning) and GCOI (Coordinating Group for Interconnected Operation). These twogroups perform stability studies, and establish recommendations concerning the requiredcontrol actions for system stabilization.

New scenarios for the Brazilian electric system. The last Ten-Year Plan, released everyyear by GCPS, estimates that the rise in electrical energy demand in the period 1997–2006 will call for the installation of an additional 3200 MW of generation every year.Two immediate questions appear: a) How will the transmission system evolve? and b)What will be the expansion process for this additionally needed generation?

Taking into account the ongoing restructuring process of the Brazilian electrical industry,the government stimulus to private investors, and the highly developed technology forcombustion turbines, it’s possible to envision the following scenario:

• Significant increase in thermal generation, mainly gas turbines. It’s expected that inthe next ten years gas turbines will represent 10% of the total installed capacity.

• Implementation of several international interconnections, initially with Argentina,Uruguay and Bolivia, through long distance or back-to-back HVDC links.

• Utilization of alternative energy sources: Wind power and solar generation, biomass(sugarcane leftovers), mainly in the northeastern part of the country.

• Significant rise in distributed generation.

• Operation of the existing system closer to its maximum limits.

The above scenario is very likely because:

• The country needs to increase its generation capacity in the immediate and nearfuture, so as to prevent severe power shortages.

• The newly implemented legislation regarding IPPs and the open access createdfavorable conditions for these new agents. The reduced construction period for gasturbine power stations is ideal for rapidly commissioning the needed additionalgeneration. Another advantage is that gas turbines can be located close to the loadcenters, therefore minimizing high investments in long distance transmission.

• The technology development of combustion-turbine driven power plants makes thecombined-cycle power plant one of the most efficient forms of power generation,offering very-competitive energy prices. These power plants cause low environmentalimpact, with negligible levels of audible noise, atmospheric pollution and emission ofliquid or solid waste. The turbine has acoustic insulation, the natural gas has very lowlevels of sulfur, and the burners can meet the most severe environmental legislation.The gas supply to these power plants is guaranteed by Petrobras (newly-discovered

8-4

gas fields as well as imported gas), together with the gas pipelines Bolivia–Brazil andArgentina–Brazil. There will also be gas pipelines within the Brazil to distribute gas.

• Another important factor is the co-generator with Petrobras as the biggest.Considering the various oil fields and refineries owned by Petrobras where the gas iscurrently being burned, it’s estimated that as much as 10,000 MW can be generated.

• The north–south interconnection, commissioned February 1999, interconnecting thenorth/northeast system to the larger south/southeast/centerwest system. It’s a 500-kVcircuit, 1,000 km long, which will bring a gain of 600 MW of guaranteed energy bymaking optimal use of the hydrological diversity between the river basins involved.This interconnection together with those with Argentina will cause some areas of thesystem to operate close to their maximum transmission capacity. As described inChapter 7, this interconnection has TCSCs for enhancing stability [8-6].

The scenario calls for the solution of an important structural problem: How to stimulateprivate agents to build hydro plants? There is still a considerable hydro potential to beexplored in Brazil that is economically feasible. This requires, however, an intensivecapital investment, and private agents prefer investments that can be recovered in shorterperiods. Undoubtedly, the solution for this problem is to form partnerships betweenprivate investors and Eletrobras, so as to ensure that investments that are sound for thewhole system will actually be made. In this case, long distance transmission with the needfor stabilization actions will be required.

In addition to the above factors and uncertainties that have a major impact on the overallsystem dynamic performance, there are the effects of major changes in the generationscenarios over the next two to three years. A significant amount of gas-fired generationwill be operating close to the major load centers.

All of the above factors point to continued importance of the phenomena of systemstability and increasing dependence on control actions. The problem in the newcompetitive generation framework is how to establish costs, both first costs and operatingcosts for controls and how to allocate them among the various parties.

8.1.2 The Nordel power systemThe Nordel power system comprises the interconnected power systems of Norway,Sweden, Finland, and part of Denmark. The other part of the Denmark, which isinterconnected with the UCPTE system, has strong connections to Nordel through severalHVDC links. The Nordel system has undergone major changes during the last decade dueto restructuring. The aim of this section is to give a brief overview of the changes, and todescribe possible impacts with respect to stability control.

The Nordic power systems are characterized by a mix of hydro and thermal generation.While Norway has almost 100% hydro generation, Finland has mainly thermal generationand Sweden has an even mix of thermal and hydro generation. The Danish system isunique with a high penetration of wind energy and co-generation from independentproducers (see next section).

8-5

Fig. 8-1. Generation and transmission capacities in the Nordic Countries.

Fig. 8-1 shows the total annual generation in the Nordic countries, and the transmissioncapacities between the countries and to the European continent.

Restructuring in the Nordic power systems. The restructuring started in 1991 withderegulation of the Norwegian electricity market. Sweden followed in 1996, and Finlandjoined the common Nordic market in 1998.

Restructuring, in general, deals with the following issues:

• Unbundling of services.

• Deregulation within trade of electrical energy.

Germany

U C P T E

600 MW1400 MW

1000 MW

630 MW

670 MW

1050 MW

1600 MW

500 MW

500 MW

200 MW

700 MW900 MW

740 MW

100 MW

HydroNuclear

Fossil

TWh80

60

40

20

0

TWh120

100

6020

0

80

20

30

TWh605040

20

010

TWh80

60

40

20

0

50 MW

1100 MW

Finland

1800 MW(future)

Sweden

Norway

600 MW

PolandHolland

600 MW(future)

N O R D E L

Russia

Denmark

8-6

• Dis-aggregation of utilities. Economically and functionally separate units areestablished within power generation, transmission and distribution, power marketsand retail sales.

In the Norwegian case, the major arguments for restructuring of the electricity markethave been to:

• Avoid excessive investment.

• Improve selection of investment projects.

• Create incentives for cost reduction.

• Create equity among consumers.

• Achieve reasonable geographical variations.

The main actors in the Norwegian (and Nordic) power market are:

Regulator. The Regulator of the Norwegian power industry is the governmental body“The Norwegian Water Resources and Energy Directorate” (NVE). The Regulator grantsregional concessions and concessions for trade in electrical energy and has an importantrole in supervision of the monopoly operations in transmission and distribution.

Market Operator. The Market Operator is responsible for the market clearing process inwhat is called the organised markets. The operator of the common Norwegian/Swedish/Finish market is Nord Pool. Nord Pool is also open to market participants withoutphysical access to the Nordel grid. See Nord Pool’s web site [8-17].

System Operator. The main grid company, Statnett SF, has the system operatorresponsibility in Norway. Similarly, there are independent system operators in Swedenand Finland, Svenska Kraftnät and Fingrid, respectively [8-17]. These companies are alsothe main transmission grid owners in their respective countries.

Market Participants. The Market Participants are buyers and sellers in the market, andinclude generators, distributors, industry, and traders/brokers.

Network Owners. The Network Owners have by regulation been given the responsibilityfor generating and distributing metering and settlement data, and keeping continuoustrack of the information so that equal opportunities are given to all the competitors.

The report Deregulation of the Nordic Power Market, Implementation and Experiences1991–1997 [8-18], issued by SINTEF Energy Research, Statnett, Nord Pool and theNorwegian Electric Federation, provides further information.

Retail Sales. Retail sales are yet another service made possible through deregulation, butis only indirectly related to the power exchange. Retail sales mean that the individualelectricity consumers are free to choose from which power company they buy theirenergy, totally independent of which network owner (distribution grid) they are connectedto.

Changes in system operation from restructuring. There are some major changes fromsystem restructuring that affect system operation and control. These relate to changes in

8-7

objectives, responsibilities and ownership, as well as to new services and ways to operatethe system.

Changes in responsibilities and ownership. This has to do with the unbundling ofservices that defines the responsibilities and tasks of the different entities. There is a mixof power producers, which are mainly economically motivated. Their main controlobjectives relate to control and optimization of their own energy resources and marketobligations. System responsibilities, such as stability control, contribution to activereserves/frequency control, and reactive reserves/voltage control become secondaryobjectives.

Increasing focus on cost efficiency. This relates to both operation and to changingattitudes toward investments in new generation and transmission capacity. A result is thatfewer new lines are being built, and the systems are operated closer to their capacitylimits. New controls rather than new transmission lines will increasingly solvetransmission congestion.

Changes in operating patterns. Deregulation of energy markets and increasingcompetition among the power producers lead to larger and more frequent changes inpower flow patterns.

New services are introduced to deal with the changes discussed above. Monitoring andcontrolling system stability, system reserves, transfer limits, etc., which are the mainresponsibilities of the system operator, is to a large extent based on ancillary services.

Ancillary services are fundamental services needed in order to maintain acceptable powerquality and power system security. The system operator will normally contract or requirethe individual power producers to provide some system services. The services may rangefrom primary frequency and voltage control, including stabilizing control, provision ofactive and reactive reserves and system protection (load shedding or generator tripping)schemes. Ancillary services can be organized as firm requirements (e.g., primarycontrols), possibly with fixed economic compensation, as contracted services (bilateralcontracts between the system operator and a power producer) or as market-based services.Secondary controls for congestion management or power balancing may also be definedas ancillary services, but organized in the Scandinavian countries through separatemarkets.

Impact on system stability and control. The changes from restructuring may impactpower system stability and system controls.

Experience indicates that deregulation has caused decreasing investments in newtransmission and generation capacity, and thus the existing systems will be operatedcloser to their capacity limits. Increased utilization means less reserves and moretransmission congestion. Thus the need for stability controls will also increase. Largerand more frequent changes in power flow patterns will increase the need for coordinatedand more robust control solutions.

Well-functioning system (or ancillary) services are crucial in the restructuredenvironment. In order to become less dependent on ancillary services provided by

8-8

generators, it is likely that system operators will show increasing interest in deployingpower electronic devices for congestion management (power flow control) and stabilitycontrol. Development and application of new energy storage devices for fast-actingreserves may also become more attractive in the future.

Another way of handling transmission congestion is to rely more heavily on specialprotection schemes. Thus there is a need for robust and coordinated design in order toavoid adverse interaction between protection systems and other controls.

In order to monitor and coordinate the increasing control applications, there is a need forimproved EMS tools at the system -control centers. On-line tools for voltage stability andtransient stability assessments will become increasingly important.

In summary, the major impact on the technical side from system restructuring is anincreasing dependence on controls in order to cope with the increasing competitionamong power producers and the increasing utilization of existing transmission grids. Thisdependence on both existing and partly new control devices will require sophisticateddesign as well as improved tools for on-line system operation.

Power system security and power quality may be regarded as collective benefits. Inderegulated systems the system operator is given the overall responsibility formaintaining the security and quality criteria. Having one such independent entity mayalso prove advantageous regarding the technical possibilities of providing coordinatedcontrols.

8.1.3 The Danish electric systemLarge variations in power transfers in transmission networks of large interconnectedsystems must be expected in the future. One part of the power variations will come fromcontrolled power plants delivering power to remote consumers or power companies onshort-term conditions. For this type of power variations, the system can at least have ashort time warning, and the system operation can be adjusted to be able to handle thepower transfer.

Another part of the power variations will come from uncontrolled power plants such aswind generation. It will be most easy to include a large amount of wind power in a system ifthe natural changes in the power production is allowed to spread freely over a large area.The probability that all the wind production will change rapidly in an equal way is smallerthe larger the system is. By allowing the power variations to spread freely over the entiresystem less demands will be put on the control of the controllable power plants in order tomaintain a local power balance. Besides, wind power and hydro power combines very well,and a free exchange of power between hydro areas and wind areas is desirable even whenlocated far from each other.

In Denmark 800 MW of wind power was installed in 1997, out of a total capacity of10,000 MW.

Both the desire for larger power transfers and the increased uncertainty of the powerchanges create a need for advanced angle stability control.

8-9

8.2 Coordinated Planning and Operation in a Competitive EnvironmentOrganizational and administrative issues under the new competitive environment canonly be resolved successfully following recognition of technical factors that makeinterconnected operation possible. We raise issues for discussion without advocatingparticular administrative and financial approaches.

8.2.1 Assuring compatibility of equivalent dynamic characteristicsIn the traditional vertically-integrated power company, the overall system reliability is theresponsibility of one entity, whether state or investor owned. In this environment theacceptable characteristics of generation, transmission, control and protection evolvednaturally to fairly uniform patterns among various utilities, as dictated by techno-economic considerations. There would be no tendency to under invest in one segment(generation, transmission, or controls) causing a disproportionate impact on reliability. Inthis scenario the cooperative approach to accepting one’s share of investment, dictated bydynamic considerations, was natural for mutually beneficial interconnected operation.

Where necessary, organizations such as NERC, NPCC, WSCC, ERCOT etc. in the USA,UNIPED in Europe and GCOI/GCPS in Brazil issued recommendations on practices tobe followed by all members of such power pools. Examples are in the area of primaryfrequency control (droop settings and spinning reserve) and automatic generation control(area control error reversals per hour etc.).

For interconnected systems using long distance transmission, the problem of poordamping of inter-machine and inter-area electromechanical oscillations presents a seriousreliability problem. The techno-economic solution is to distribute control effort (PSS inthis context) over most generators. Members of interconnected systems owning bothtransmission and generation follow voluntarily the guidelines set by coordinating councils(e.g., in the WSCC every unit over 75 MVA is to be equipped with a PSS).

Since this problem of damping can also be abated by adding transmission, one canappreciate the problem of enforcement of the most techno-economic solution. Thissolution is usually borne by the generation segment, where independent producers haveno perceived stake in transmission.

This dilemma extends to other system reliability aspects such as transient stability, loadshedding, and generator dropping.

Distributed generation in systems linked by EHV and UHV transmission can presentmajor challenges in system and protection design. Load rejection and system separation,with overspeeding generators connected to excessive line charging, could lead to veryhigh overvoltages and widespread damage to system and consumer equipment.

It’s not merely stability that must be addressed. The entire system design must be by ahighly trained team considering all relevant parts of the system regardless of ownership.

Reference 8-4 describes how the evolution of system structures can affect the necessityfor stabilization and its location. Loading, with its effect on angle separation and relativeinertia between sending and receiving areas, play an important role.

8-10

In systems with widespread transmission and significant interchange over long distance,the problem of oscillatory instability can dictate the need for stabilizing action undernormal operating conditions. In other systems the problem arises only followingcontingencies. Since the nature of the system structure following multiple contingenciesis almost unpredictable, units that normally are not participating in oscillation dampingaction can become important.

The effectiveness of PSS in providing damping is not only a function of their applicationon generating units, but also a function of the PSS and other excitation equipment controltuning. System-wide dependence on PSS for adequate damping performance will requiremore formal inspections and testing by the regional transmission organization (ISO orindependent transmission company) of the restructured power industry.

8.3 The Impact of IPP Thermal Generation on System DynamicPerformance

8.3.1 Beneficial aspects of IPPsNew thermal-based IPPs will bring many benefits to the interconnected system:

• Being close to the load centers they will bring better voltage control and smallerloading of transmission lines, with consequent reduction in transmission systemlosses.

• Improvements in voltage stability, because of a larger reactive power support near theload centers.

• Improvements in electromechanical stability, due to smaller line loading and addeddynamic voltage support. The damping of interarea oscillations will also tend toimprove as a function of the smaller phase angle differences.

• Extra flexibility in planning equipment and transmission line outages.

• In Brazil, the availability of more thermal generation will allow an effective hydro-thermal coordination.

• Alleviation of the problem of ever-increasing transmission distances to bring power tothe load centers.

• The better dynamic voltage control will yield a more reliable operation oftransmission line distance protection, with a significant reduction in undesiredtripping. This decreases the risk of system separation, frequency decay, and loadshedding.

8.3.2 Detrimental aspectsIPPs, when compared with state owned or regulated generating companies, are orientedtowards a higher and faster return on investment. Their motto is to maximize powerproduction and reduce their costs. What could be the consequences? Some of thefunctions carried out for free in today’s environment (voltage support, frequency

8-11

regulation, dynamic response, transient overload capability, etc.) are classified in the newenvironment as “ancillary service,” whose provision will have an associated cost. Someof these aspects are further elaborated in tables in section 8.3.4.

Offsetting some of the positive aspects of IPPs listed in §8.3.1 is the need for assuringredundancy in the case of unplanned outages of such facilities. Such outages result in lossof both power production and voltage support—which must be provided by alternatefacilities.

8.3.3 Problem issues with new IPPsTables 8-1 and 8-2 list system design and operation considerations for the restructuredindustry. These issues generally involve dynamic aspects of the plant interacting with thepower system. In the vertically-integrated traditional utility (or power pool made up ofsuch utilities), the planning and design process is usually undertaken by ownerrepresentatives participating in joint interconnected system studies with access to theentire database. Reliability criteria are followed and the design process considers alllogical cost-effective alternatives, whether they involve generation, transmission,protection or control.

In the restructured environment the technical approach should be the same since the merefact of separate ownership of generation versus transmission does not change theunderlying laws of physics which govern the reliability of overall system performance.The challenge is to develop an organizational structure to execute the necessary systemstudies and enforce the design requirements among the separate parties. As competitors,the parties have a natural tendency to hold back on free exchange of information. IPPs areconcerned with the generation process and normally would not have the expertise todetermine complex control and protection requirements dictated by the overall system.

The foregoing considerations point to the logic of a strong independent and competentorganization to not only be in charge of system operation, but also of licensing futuresystem additions in generation, transmission, control or protection.

Issues listed in the tables show the need to establish methods and procedures for requiringcertain design features in IPP installations. These include providing ancillary services—for instance reactive power support, primary speed control, supplementary dampingcontrol (PSS) etc. Little has been done so far to develop such methodology, which shouldinclude allocation of costs to those agents not contributing their share of ancillaryservices.

If this is not done in the planning process, IPP additions may impact the adequacy oftransmission networks. The resulting additional reinforcements needed in transmissionwould be reflected in transmission costs, which would have to be borne by all consumers.

8-12

Table 8-1. Protection, System Voltage/Frequency Control, and Stability Aspects

Issue Traditional Approaches Problem issues with new IPPs

Protectionsettings

Settings take into account the plantequipment’s and power systemrequirements

- Concerned only with plant equipment security.

- Larger possibility of plant tripping duringdisturbances.

- For disturbances that cause generation deficits,plant tripping will increase the magnitude offrequency dips. In this case it will be necessary toincrease the total load shedding. In extreme casesit could lead to a system collapse.

- For disturbances that cause overvoltage, planttripping can aggravate the voltage profile,increasing the chances of reaching transmissionlines protection settings. The tripping oftransmission lines could lead to a system collapse

- For disturbances that cause large absorption ofreactive power by the generators controllingvoltage profile, settings of minimum excitationlimits can cause plant tripping. With increasingnumbers of IPP plants, this can lead to largersystem overvoltage and equipment tripping (ordamage).

Generator powerfactor

Generators with low rated powerfactor (0.9) can be used. This canavoid network reinforcements.

Without consideration of system requirementsduring contingencies, the tendency would be toorder less costly, higher rated power factormachines. The deficiency of reactive powerreserve could require move expensive alternativeequipment in transmission.

Short-timeoverloadcapability ofgeneratorexcitationequipment

Exciters are able to produce up to200% of rated reactive power forapproximately 20 seconds. Thisimproves system dynamicperformance.

- Lower capacity excitation systems andconservative setting of limiters. This reduces IPPcosts. The limiter actuation might be increased toprotect excitation and generator windings againstfailure due to high voltage stress. This can leadto voltage control problems and even collapse.

- System requirements could be enhanced withgreater MVAr reserves in generators.

Operation ofgenerators assynchronouscondensers

This characteristic is used duringlight load conditions in order toprovide better voltage profilecontrol, to maintain short-circuitlevel and to avoid transmission lineopening to mitigate sustained overvoltage during light load.

This expedient would require installation ofclutches representing additional costs.

Excitationequipment,power systemstabilizers andgovernors

Are fully utilized to improve thepower system dynamicperformance, being considered themost appropriate and economicmeans.

Application of higher cost machines withimproved excitation systems (high initialresponse yielding move effective action fromPSS) would not be normally adopted withoutsome hard rules to define compensation of costs.

Participation inSpecialProtectionSchemes (SPS)

The design and implementation ofSpecial Protection Schemes(Emergency Control Schemes) areanalyzed by all parties involved.The SPS is installed consideringthe best location, i.e., it can beinstalled in any plant.

The IPP may not accept to participate in anySPS. This non-acceptance may jeopardize systemreliability and require system reinforcements.

8-13

Table 8-2. Operating Aspects

Issue Traditional Approaches Problem issues with new IPPs

Minimumnumber of unitsin operation

The number of units in each plantis determined to guarantee aminimum value of system inertiaand reserve.

- IPP could consider that they have no obligation to dothat, and maintain the number of machines in operationin order to obtain the maximum productivity of theplant.

- This could affect voltage control, system stability andincreased frequency dips.

Generating unitoperation withminor failures

When minor failures occur, theutilities may agree to keep thegenerating units in operation untilsystem conditions evolve to level atwhich unit disconnection will notjeopardize the overall systemreliability.

The IPP could consider that they have no obligationwith system reliability requirements, the main objectivebeing to protect their own equipment.

Informationexchange anddata availability

- To provide a more reliable andsecure operation, abundantinformation is made available oncurrent limitations/unavailability of equipment andpower flow constraints.

- Traditionally all the data areavailable including data fromdisturbances (oscillograms, plantoperator reports, etc.,) for postoperation analysis.

- The IPP could consider having no obligation to informthe others on what is occurring to his plant.

- This intentional withholding of information isdetrimental to overall system reliability.

- The IPP could have inadequate data acquisition andrecording equipment.

Black-startcapability

The operational planning of theinterconnected system determinesthe restoration planning with itsvarious parallel subsystems.

-Every subsystem has at least onepower station with black-startcapability.

- IPPs, for cost reduction, may rely on remote powerstation cranking rather than install black-start capability.

- As a consequence, the system restoration time may beincreased.

8.3.4 Conclusions related to IPPs1. Careful planning and design studies should establish the proper integration of IPPs

into the interconnected system, including special requirements in equipment, andcontrol and protection. The proliferation of IPPs, with their rapid installation cycle,can have detrimental impacts to system dynamic performance, which may not be fullycompensated in the transmission network at reasonable cost.

2. Although transmission network investments will rise, the overall system reliabilitycould be somewhat degraded in the future.

3. The eventual lack of control actions and the consequent rise in transmission pricescan result in loss of economic efficiency in power production.

4. Many efforts have been noted to develop methods and tools for some of the ancillaryservices. So far, however, very little has been done concerning control action costallocation. This should be considered a priority issue in order to ensure economicefficiency.

8-14

5. The transition period from the cooperative model to the competitive one will causesome additional risks, which are not fully assessed.

6. One way to minimize the detrimental impacts and additional risks, is ensure thatselling ancillary services can be good business. Something should done, so as to makegeneration fulfill its natural or traditional ancillary functions. Finding other controlalternatives in the transmission network (like FACTS) is always more expensive.

7. The Independent System Operator (ISO) concept is good, but that organization shouldhave added functions in long term operational planning and, particularly, licensingand inspection of new facilities to ensure that they meet system requirements.

8.4 Other Issues Related to Power System Performance in the NewUtility Environment

8.4.1 Reliability aspectsThe forces of market deregulation have encouraged a widespread decline in planningresources, and have undercut the planning process itself. Unrealistic models provide acommon point of failure for the entire decision making process whereby the powersystem is planned and operated. Compounding this, the system sometimes operates underconditions that planning cannot anticipate.

Market deregulation and utility restructuring are, through a variety of mechanisms,making it impossible to predict system vulnerabilities as accurately or as promptly as theincreasingly volatile market demands. Controller-based options for reinforcing the powersystem can be very attractive. For a control system to be fully competitive in this respect,however, its functional reliability must somehow be established early in the planningprocess.

It’s rarely possible to do this within the conventional framework used for newtransmission lines or for new power plants. It’ll always be necessary to trade the benefitspromised by a control system against the inevitable risks associated with closing a high-power loop around system dynamics that are not fully understood. If the risks areperceived as too high, or if the functional reliability is perceived as inadequate, thensystem reinforcements though enhanced control will be displaced by less technicallydemanding means.

Reliability is just one intangible emerging in the new power system. Others includeinformation security, regulatory changes, business survival, and the directions in which aparticular regional transmission organization (RTO) evolves.

8.4.2 Implications of equipment ownershipAs many electric power systems move toward deregulation, there is much focus on theeconomic issues associated with the new competitive operating environment; details ofenergy trading and pricing have been in the forefront. However, the ability to operate insuch an environment with an acceptable degree of security and reliability, and indeed to

8-15

be economically competitive, requires significant attention to the methods and strategiesof power system control.

In the new environment, the power system comprises corporate entities having diverseroles, equipment, and business interests. There are independent generating entities,transmission entities, distribution entities and brokering entities. The physical functioningof the integrated power system, however, remains the same as before. Therefore, theresponsibility for control of individual equipment should not follow ownership; instead itshould be vested with RTO. The specification and design of these controls should be partof overall system planning and design carried out by an independent entity. Otherwise,system security and economy will be sacrificed, defeating the very purpose ofrestructuring the industry.

In particular, it’s essential to recognize the critical role played by generator controls inmaintaining system stability and controlling voltages and frequency. It should bemandatory for the generators to be fitted with fast-acting excitation system, AVR, andspeed governing systems. In many cases, PSS should be mandatory.

The PSS should be designed and tuned so as to contribute to the enhancement of overallsystem stability, including damping of local as well as interarea modes of oscillation.

There should be no difficulty in motivating power plant owners to install controls thatenhance the operability and stability of the generators. For those controls that areprovided to meet the overall power system requirements there should be proper financialincentives.

8.4.3 AGC in the new environmentWith deregulation comes the redefinition of system control areas. Both the introduction ofnew control areas and the consolidation of existing controls areas impact the waytraditional control issues are handled. Traditionally, frequency control, achieved throughthe matching of generation to load, has been one of the functions of control areas usingsome form of automatic generation control (AGC). Although the extent to whichfrequency control is required is debatable, some control is required to prevent theinstabilities and other adverse effects associated with excessively low or high frequencies.The control of frequency to tight tolerances is arguably associated with improved powerquality which may be expected by some customers, but that is not strictly a requirementfor successful interconnected operation.

In a deregulated environment it’s not clear who will be responsible for any level offrequency control. AGC requires spinning reserve that can be valued as an ancillaryservice. While it’s possible that certain parties will be prepared to provide such servicesfor a price, what is not clear is the extent to which this will occur. The first issue ofmaintaining the frequency of a large system within limits required for secure operation isa natural byproduct of near matching of the load and generation which should take placeunder free energy trading.

The second issue of maintaining tight frequency control for power quality concernsshould be based on value and price to consumers.

8-16

8.4.4 Modeling/data requirements — a bigger challengeEqually important as the analysis method is the model used to represent the powersystem. It’s essential the model represent sufficient detail and accuracy to properlyreproduce all important system dynamics. While this has led to the use of very largesystem models (for example, North American Eastern Interconnection is oftenrepresented by more than 26,000 buses), analytical tools are available to handle suchsystems. Good dynamic reduction methods are also available which can be applied toreduce large models to more manageable sizes while retaining the key system dynamics.

Perhaps a bigger challenge is the availability of model data for various equipment,including generators and the associated controls, protective systems, and system loads.While phenomenal advancements have been made in terms of analytical techniques andcomputational tools, data acquisition has not kept pace with the requirements. Manyutilities use “typical data” for modeling much of the equipment. For control andprotection, the data is often not representative of the actual settings and, in many cases,the condition of the equipment. More effort is needed towards the acquisition andverification of model data. This is being increasingly recognized by the industry,particularly in the aftermath of major system disturbances. For example the twodisturbances that occurred in 1996 on the western North American system have motivatedthe WSCC to mandate field measurement and model derivation for all generation units(unit, exciter, PSS, governor, and protection) greater than 10 MVA. Once good modelsare obtained (that is, they match the field response), then it’s necessary to use thisinformation to optimally tune the system. Once optimized, it’s essential that fieldadjustments are not permitted without prior study of the impacts.

8.4.5 On-line dynamic security assessment and real-time monitoring andcontrol

In the new power sector the system conditions are extremely unpredictable and thevolume of transactions that may have to be examined may be huge. The traditionalapproach to deploying preventive and emergency controls based on off-line securityanalysis studies which generate a set of tables indicating stability limits and controlmeasures may not be satisfactory.

In the new structure, tools are necessary, such as on-line transient stability assessment andvoltage stability assessment. These are described in Chapter 5.

In order to make these new tools useful, it’s necessary obtain reliable on-line input data.

It’s necessary adopt real-time system monitoring and control. An example of such ascheme is a wide-area measurement system (WAMS) being developed by BonnevillePower Administration western North American. The WAMS use synchronized phasormeasurements and portable power system monitors to centralize information at controlcenters.

8-17

8.4.6 Alternatives for pricing of stability controls in a deregulated industryThe shift to a market-based structure necessitates the unbundling of services by strippingout non-energy costs and identifying ancillary services that have costs and value. Thefollowing is based on procedures of the Northeast Power Coordinating Council in NorthAmerica [8.3].

With the re-regulation of the electric power industry one important question appears: Willproper market signals in combination with commonly accepted “best practices” fostercompetition and preserve or even enhance the reliability of the system? This represents adifficult challenge with respect to the interface between the market driven generationsector and the regulated transmission system that may be under the control of anIndependent System Operator (ISO). However, the prudent use of Special ProtectionSystems (SPS) and Dynamic Control Systems (DCS) can play a vital role in enhancingboth competition and system reliability provided that proper market signals areimplemented.

Performance requirements. In all cases of SPS, the design and operation must beconsistent with all criteria, including protection criteria. Depending upon the type of SPS,varying degrees of functional redundancy may be required to ensure reliable operation.For example, Type I (SPS with potential for interarea impact, initiated by normalconditions) may require two independent protection schemes while a Type III (SPS withpotential for local impact only) may require only one set of system protection. In addition,for loss of an element without a fault or due to a single line to ground fault cleared innormal time, the failure of an SPS circuit breaker is considered as part of the normalcriteria. Thus there are situations where excess generation may be armed for rejection toensure that sufficient generation is successfully tripped for a critical fault.

The design and operation of the DCS must be approved by the ISO. Once approved,procedures must ensure that the DCS performs as intended. Note that the NERCStandards require the generator owners to provide accurate and timely steady state anddynamic data for their generating units [8-8]. Modeling should be consistent with industrystandards, such as IEEE models. In order to ensure the proper modeling of excitationequipment (also other machine and governor parameters), the ISO could conduct audits(similar to machine parameter measurement R&D projects) as required. In addition, eventreconstruction by simulating actual system events and comparing the results with theactual machine performance could identify units with suspect parameters. It’s necessarythat any changes to the control parameters be communicated to the ISO.

DCS are subject to reliability standards that ensure dependability and security. For Type IDCS (whose incorrect operation or failure to operate following a normal criteriacontingency would have interarea or interregional consequences), design requirementsspecify that the DCS should perform its intended function for specified Bulk PowerSystem (BPS) contingencies while itself experiencing a single undetected failure. Thismeans that vital subsystems should either have a functional redundancy or sufficient self-diagnostics so that there would be reduced dependency on the DCS in settingtransmission system limits. All Type I and Type II DCS (installed for the purpose ofmitigating the interarea impact of extreme contingency) are designed so that a critical

8-18

failure of the DCS itself does not cause unacceptable BPS behavior. Similar to protectionsystem criteria, owners of DCSs have obligations to perform both maintenance andmonitoring functions.

Justification for SPS or DCS. The implementation of a SPS or DCS is dependent uponthe system conditions that justify their use. We discuss two main categories of SPS use:reliability and economy.

Reliability. An approach to defining reliability is to recognize that the SPS or DCS isproviding greater resiliency to the operation of the network when the device is notrequired in the setting of normal limits on the system. For all DCS and those SPSrequired for stability, the devices are in effect providing greater stability margin to thesystem for a particular set of contingencies. An alternate approach to reliability does notaccount for the additional robustness of the system, but rather defines reliability as therequirement that the implementation of a DCS or SPS cannot reduce the operating limitsof the network. (If there is a reduction, then there are economic penalties.)

If the system is operating in an insecure state (for either normal or extreme contingencycriteria) and the arming of an SPS or DCS would return the system to a more secure state,the SPS or DCS becomes essential to maintaining reliability. For example, immediatelyupon the loss of one or more transmission facilities, the interface flows may violate thepermissible normal criteria transfer limit. For this scenario, a Type I SPS could restore thetransfer limit.

Economy. The economic use of an SPS or DCS applies when the device is required toincrease the normal transfer limit of the system. In this case, the use an existing SPS orDCS as well as the planning of a future SPS or DCS would be driven by the transmissiontariff structure.

In the future it may be possible to attribute an improved loss of load probability toparticular SPS or DCS. This could then be weighed against the value that the load placeson enhanced reliability of service. It is judged that this will present not only technicalchallenges, but will no doubt be complicated by the regulatory process required toapprove this methodology.

Payment Schedules for SPS and DCS. Several options exist for the payment schedulesfor the arming of existing SPS and DCS, as well as for the implementation of futuredevices. Some options:

Don’t Pay. In the Don’t Pay scenario, it’s assumed that all existing SPS and DCScontinue to function in a secure manner, but there is no special payment made for theiruse. Future system additions could be addressed through rules such as a requirement thatall future generators are required to have high performance excitation systems thatinclude power system stabilizers (PSS). The Don’t Pay method could also requirepayment from the SPS or DCS providers if they failed to preserve existing system transferlimits.

It’s not clear if the Don’t Pay option will cause any degradation in the reliability of thesystem with respect to the implementation of SPS and DCS. In the short term the primary

8-19

focus of generator providers will be on issues that are more economically lucrative. It’salso well recognized by market participants that the great experiment in the deregulationof the electric power industry could come to an abrupt end if there were manyinterruptions of load. In the long term, the robustness of the network could improve ifprice signals locate new generation closer to the load and overall system transfers arereduced. This scenario would of course reduce dependency on SPS and DCS.

Embedded Cost. This method recognizes the benefits of SPS and DCS and seeks to makethe provider cost neutral, but without necessarily accounting for lost opportunity costs.We discuss the possible use of an Embedded Cost method, first for SPS and then DCS.

The payment for arming a SPS would include paying for the installation and maintenanceof the protection system as part of the Transmission Service Charge (TSC) orTransmission Uplift Charge (TUC). For Type I SPS the ISO and all market participants(the generator, transmission owner, and the load) are the beneficiaries of the reliabilityaspects of the SPS. However, the generator is not compensated for any additionaltransmission capability that may be available as the result of arming or installing the SPS.For Type II SPS the system has an extra degree of security against extreme contingencies.However, the generator, transmission facility, or load providing this service is placed atrisk. In the event that either Type I or Type II SPS is triggered and works as designed foran actual contingency or has an undesired trip (within reason), payment shall be asfollows:

GR (generation rejection or reduction)—The unit must be made “whole” otherwise theunit would not be willing to provide the extra measure of security. Therefore, back-uppower is supplied free of charge to the generator if it is rejected. (It’s assumed that only alimited number of false trips due to the SPS would be tolerated.) If this power is from theeconomy (perhaps the Location Based Marginal Pricing or LBMP) market, then thedifferences between the economy market and the rejected generator’s price is providedfrom the TCT (Transmission Cross-Tripping).

TCT (Transmission Cross-Tripping)—No payment. The ISO has responsibility for systemreliability and the TCT provides an extra level of security.

LR (load rejection)—No payment. SPS is in the same class as underfrequency loadshedding. Eventually there may be reliability-based rates and the load which is placed atrisk might get a discount.

For Type III SPS (with potential for local impact only)—the local area is the beneficiaryof the SPS. Local arrangements could be made to compensate the involved parties.

A proposal for the payment of the DCS is dependent upon several factors, includingwhether the control is excitation equipment or governor related, the type, and the inherenttransmission rate structure.

Excitation equipment tuning and supplementary controls, such as power systemstabilizers, are essential to the stability performance of the system. They are considered aspart of the “Voltage Support and Control” Ancillary Service and the payment for thiscategory of DCS is thus highly dependent upon the transmission rate structure.

8-20

Transmission tariffs for Voltage Support and Control are often “embedded” or cost basedrates. For this scenario, the generator is paid for a portion or the full capital and operatingcost of the DCS.

Excitation equipment improvements, such as replacement with solid state systems and/orthe addition of PSS could result in greater system resiliency, particularly with respect toextreme contingencies. In other instances, the modification to the excitation system maynot be capital intensive and could require simply changing a gain. In either case it’ssuggested that the embedded cost method would pay the generator for all or part of theexcitation equipment modification. It’s recognized that the generator would not realizeany additional benefits from increases in transfer limits.

Turbine governor DCS fall into several categories. Those that provide frequency responseand regulation services usually impact the long-term dynamics of the network and arecommonly addressed by transmission tariffs. It would be a difficult task to economicallyquantify the differences between the control performance (AGC) response and the DCSturbine governor response. Other DCS, such as fast valving, impact the short-termdynamics and can be handled similar to a SPS.

Market-Based Rates. The proper price signals would establish an economical incentivefor providing SPS and DCS services. It’s interesting that generators on the downside of,and loads on the upside of, a congested transmission interface might be reluctant to makesystem improvements resulting in higher transmission operating limits. This is becausethe higher limits would result in the generator receiving a lower LBMP and the loadpaying a higher LBMP price. The generator could proceed with the improvement bymaking financial arrangements with other market participants. However, the question ofhow the ISO would arrange for system improvements justified by improved systemresiliency needs to be determined based upon the particular ISO definition or reliability.We now discuss possible Market-Based Rate methodologies for SPS and DCS.

For a Type I SPS resulting in higher system transfer levels, there is a cost saving tocustomers. However, the SPS comes at a cost to the owner of SPS, possibly including alost opportunity cost, as would be the case for the rejection of generating units. The SPSholder could theoretically claim an allocation of transmission that could be handled in anynumber of different ways. The transmission allocation could be defined as the increase intransfer capability across a congested interface. In this case, the Transmission Providercould offer a payment based upon a percentage of the expected increases in wheelingrevenue. An interesting approach to determining the value of and location of thetransmission allocation could be the auction of incrementally feasible TransmissionCongestion Contracts (TCCs) as suggested in some Locational Based Marginal Pricing(LBMP) methods. The auction could be conducted similar to proposals for the conversionof “traditional” transmission rights into TCCs. Alternatively, it may be possible for theSPS holder to utilize a more direct bid based methodology that avoids the complication ofa special TCC auction as suggested by some LBMP systems. The following alternativebid based system could be used:

GR—Since generation rejection schemes allow for higher economy transfers by placingunits at risk, the benefiting entities should pay the machines for accepting a possibly

8-21

lower capacity factor. Other costs to the machine include possible penalties for backupsupply (assuming a bilateral contract) and physical costs, such as possible loss of life ofthe machine from additional trips. The generator would need to weigh these costs againstpossible lost opportunity costs due to the lower transmission interface capability thatwould result from disarming the SPS. It’s suggested that the following procedure beinvoked:

The ISO determines transfer limits with and without the SPS activated.

The generation unit (through a power exchange) accepts or rejects bids from othergenerating units or load serving entities for the activation of the SPS.

TCT—Presumably higher system transfer limits would result in greater transmissionrevenues. Therefore, the owner of the TCT is reimbursed for the costs of the SPS throughthe Transmission Uplift Charge.

LR—Load rejection is the dual of generation rejection and could be handled a similarway as follows:

The ISO determines transfer limits with and without the SPS activated.

The Load (possibly through a power exchange) accepts or rejects bids from othergenerating units or load serving entities for the activation of the SPS.

Type I SPS do not increase transfer limits and Type II SPS are reliability based and do nothave an economy market at this time. In the future, it’s conceivable that loads may wishto pay for higher levels of reliability. At that time a power exchange could be used as amechanism for bidding for the activation of the SPS. It’s envisioned that alternativeapproach would be an ISO calculation of the probability of the contingency eventsnecessitating the SPS action. This could be weighed against the cost of the serviceinterruption. Based upon this economic determination, a decision could be made onwhether or not to pay the reliability-based rate for the SPS.

Type III SPS is a local issue where it is difficult to generalize reliability versus economymethod of compensation.

For the case where the transmission system is stability limited, the application of a singleDCS could increase the transfer capability of the system. Similar to the methodsdescribed in the SPS section, increased transmission capability could be allocated to theowner of the DCS. This method is applicable to new or improved DCS as well as forDCS that can be armed or disarmed by operators or a defined set of system conditions.The transmission allocation problem becomes more complex for the case where multipleDCS are coordinated to increase transfer limits. Here the individual owners of the DCScould come to some business solution, possible based upon techniques that are used fortuning DCS and Dynamic Security Analysis.

Total payment for DCS used to enhance the reliability of the network would bedetermined similar to the method used for SPS. However, the allocation of the paymentwould be more complicated and possibly require advanced analyses that determine theindividual contributions of the DCS

8-22

Market Power. In all cases where the owner of an SPS or DCS is paid there is the issueof market power. It will be necessary to prevent anti-competitive actions by generatorsand loads by constant observation and possible dispute resolution by regulatoryauthorities.

8.4.7 Large scale stability controls and legal liabilitiesLegal liability with stability controls may be a concern when the nature and role of theRTO (Independent System Operator or Independent Transmission Company) is not wellestablished [8-14–16]. Key points are:

• Redefinition of electricity as a market product may well expose all providers to legalliabilities from which they are now immune.

• Large-scale stability control (LSSC) faces many technical challenges that make it verydifficult to assure reliable LSSC performance. Only the RTO(s) will have theinfrastructure and other assets needed to monitor LSSC performance effectively.LSSC, marketed as an ancillary service, could be a magnet for lawsuits. An RTOshould be held harmless for duties performed according to sound engineeringpractice,

• LSSC actions that are initiated after system failure is clearly underway face less legalexposure. This would favor a shift in emphasis, toward greater acceptance of systemfailures but with LSSC action to make the failures “graceful” and to facilitate promptrestoration of electrical services.

References and bibliography8-1 M. K. Donnelly, J. E. Dagle, D. J. Trudnowski, G. J. Rogers “Impacts of the

Distributed Utility on Transmission System Stability,” IEEE Transactions onPower Systems, Vol.11, No. 2, pp. 741–746, May 1998.

8-2 H. Clark, PTI Newsletter, Issue No. 87, Fourth Quarter 1996.

8-3 M. Henderson, “Stability Controls in a Restructured Industry,” presentation toIEEE/PES Power System Stability Controls Subcommittee, January 1997.

8-4 F. P. de Mello, “Some Aspects of Transmission System Planning and Design inDeveloping Countries,” Engineering Foundation Conference, Henniker, NewHampshire, Aug 21–27, 1976.

8-5 M. M. Adibi and L. H. Fink, “Power System Restoration Planning,” IEEETransactions on Power Systems, Vol.9, No. 1, pp. 22–28, February 1994.

8-6 C. Gama, R. L. Leoni, J. Gribel, R. Fraga, M.J. Eiras, W. Ping, A. Ricardo, J.Cavalcanti, and R. Tenório, “Brazilian North–South Interconnection —Application of Thyristor Controlled Series Compensation (TCSC) to Damp Inter-Area Oscillation Mode,” CIGRÉ, paper 14-101, 1998.

8-7 D. Shirmohammadi and A. Vojdani, “An Overview of Ancillary Services,” VSEPOPE, May 1996, Recife, Brazil.

8-23

8-8 North American Electric Reliability Council, NERC Planning Standards,September 1997, www.nerc.com.

8-9 P. Kundur, Power System Stability and Control, McGraw-Hill, 1994.

8-10 P. Kundur, M. Klein, G. J. Rogers, and M. S. Zymno, “Application of PowerSystem Stabilizers for Enhancement of Overall System Stability,” IEEETransactions on Power Systems, Vol.4, No. 2, pp. 614–626, May 1989.

8-11 L. R. Day, “Control Area Trends: Principles and Responses,” IEEE ComputerApplications in Power, pp. 34–39, April 1995.

8-12 L. H. Fink, P. J. M. von Son, “On System Control within a RestructuredIndustry,” IEEE Transactions on Power Systems, Vol.13, No. 2, pp. 611–616,May 1998.

8-13 P. Kundur and G. K. Morison, “Power System Control: Requirements and Trendsin the New Utility Environment,” Bulk Power System Dynamics and Control IV -Restructuring, August 24–28, Santorini, Greece.

8-14 J. F. Hauer and C. W. Taylor, “Information, Reliability, and Control in the NewPower System,” Proceedings of 1998 American Control Conference, June 24–26,Philadelphia, Pennsylvania.

8-15 A. J. Roman, “Legal Responsibility for Reliability in the New CompetitiveElectricity Markets in Canada: Who Do I Sue if the Lights Go Out?” plenarysession of the IEEE/PES Summer Meeting, Edmonton, Canada, July 18–22, 1999(to appear in IEEE PES Review).

8-16 B. J. Fleishman, “Emerging Liability Issues for the New Electric Power Industry,”1997 IBC Conference on Ensuring Electric Power Reliability in the CompetitiveMarketplace, San Francisco, September 29–30, 1997.

8-17 Internet addresses: http://www.nordpool.no/. See also: http://www.statnett.no/,http://www.svk.se/ and http://www.fingrid.fi/.

8-18 Deregulation of the Nordic Power Market. Implementation and Experiences1991–1997. SINTEF Energy Research, Statnett, Nord Pool, Norwegian ElectricFederation 1997 (http://www.energy.sintef.no/publ/rapport/97/tr4602.htm).

Chapter 9

Conclusions and Suggested Future Work

Angle stability control is an old power system problem, with many effective solutions.The present deregulated, competitive environment for the generation subsystem, however,presents new challenges; power transactions may be very different than planned, withneed to increase stability-related transfer limits. New long-distance interconnections onseveral continents present synchronous stability challenges. We wish to exploit recent andemerging technologies for the development of cost-effective advanced stability controls.

Technologies include high voltage power electronics, and the various informationtechnologies such as digital sensors and signal processing, digital controls, digitalcommunications, fiber optics communications, GPS, intelligent controls, and advancedcontrol theory.

Questions investigated by the task force include:

• What is the value and application of wide-area (centralized) stability control?

• What is the value and application of direct control of rotor angles?

• What are needs for adaptive control?

• What new control techniques (examples: robust control theory, fuzzy logic) arepromising?

9.1 Conclusions1. The primary stability controls are fast fault clearing and generator excitation control.

Special feedforward controls such as generator tripping for severe disturbances arevery effective and are widely used.

2. Generator excitation control and control of other existing actuators should be fullyexploited before considering transmission level mechanically-switched or powerelectronic controlled equipment.

3. The purpose of stability controls is to remove stability-imposed limits on powertransfer. High damping ratio for oscillation damping or “stiff” (high synchronizingpower) performance may not be cost-effective. Direct control of rotor angle is notnormally appropriate.

4. For cost and reliability/complexity reasons, local control strategies are the first choice.Control and communication technologies allow wide-area control where benefits(e.g., superior observability) exist.

5. Digital controls should not be simple replicas of analog controls. Possibilities forcontrol adaptation, control mode shifting, and different control structures should beconsidered.

9-2

6. Time and frequency domain simulations are essential for robust stability controldesign and for control certification. This requires development of accurate models anddata sets. Simulations must include sensitivity analysis of various operating/disturbance conditions and other uncertainties. Simulations should be validated byfield tests and system monitoring.

7. Wide-area monitoring of power plants and substations is desirable to support stabilitycontrol implementation and operation.

8. Control reliability should not be based on simple redundancy requirements. Rather,hardware and software algorithm failure modes and frequency should be investigated,along with the consequences of failures.

9. With independent ownership of generation, requirements to maintain stability ofsynchronous generators remain. Overall power system engineering for stability isrequired. Some system requirements should be mandated. One example is generatorautomatic voltage regulation. Other requirements are suitable for ancillary servicearrangements.

10. Transmission-level power electronic equipment offers many possibilities for powerfulstability control. These are available for special needs, and ongoing development maymake the equipment cost-effective for more widespread use.

11. Synergies are possible between stability control and control center EMS (energymanagement system) applications. Dynamic security assessment may be used forcontrol arming and adaptation, or as the database for pattern-recognition basedcontrols.

12. “Defense-in-depth” and “multiple lines of defense” are essential to minimizecatastrophic power system instability and widespread outages because of rare multipleoutages and failures. Stability controls may include load shedding and controlledseparation. Power plants should be able to withstand voltage and frequencyexcursions associated with islanding and other abnormal conditions.

9.2 Areas for Future Work1. Wide-area control based on new communication technologies. Digital fiber optic

communication is rapidly becoming available. Emerging technologies such as lowearth orbit satellites are promising. Direct load control is facilitated by information-age technology.

2. Further exploitation of digital control possibilities that break paradigms establishedduring the decades of analog control development.

3. Modulation of steam and gas turbines mechanical power for damping of lowfrequency oscillations.

4. Integration of control center data/application programs with stability controls. Aparticular challenge for on-line interarea stability assessment is state estimation forpower systems spanning large portions of a continent.

9-3

5. Strategies and criteria for stability control in the partially deregulated and restructuredelectric power industry. This will include better-defined mandatory practices withenforcement, and also ancillary service markets for power system stability enhancingcontrols.