microgrids: resilience, reliability, and market structure
TRANSCRIPT
Southern Methodist University Southern Methodist University
SMU Scholar SMU Scholar
Electrical Engineering Theses and Dissertations Electrical Engineering
Spring 5-2018
Microgrids: Resilience, Reliability, and Market Structure Microgrids: Resilience, Reliability, and Market Structure
Saeed Dehghan Manshadi Southern Methodist University, [email protected]
Follow this and additional works at: https://scholar.smu.edu/engineering_electrical_etds
Part of the Power and Energy Commons
Recommended Citation Recommended Citation Dehghan Manshadi, Saeed, "Microgrids: Resilience, Reliability, and Market Structure" (2018). Electrical Engineering Theses and Dissertations. 8. https://scholar.smu.edu/engineering_electrical_etds/8
This Dissertation is brought to you for free and open access by the Electrical Engineering at SMU Scholar. It has been accepted for inclusion in Electrical Engineering Theses and Dissertations by an authorized administrator of SMU Scholar. For more information, please visit http://digitalrepository.smu.edu.
1
MICROGRIDS: RESILIENCE, RELIABILITY, AND MARKET STRUCTURE
Approved by:
Dr. Mohammad Khodayar
Dr. Jianhui Wang
Dr. Behrouz Peikari
Dr. Khaled Abdelghany
Dr. Eli Olinick
1
2
MICROGRIDS: RESILIENCE, RELIABILITY, AND MARKET STRUCTURE
A Dissertation Presented to the Graduate Faculty of the
Bobby B. Lyle School of Engineering
Southern Methodist University
in
Partial Fulfillment of the Requirements
for the degree of
Doctor of Philosophy
with a
Major in Electrical Engineering
by
Saeed Dehghan Manshadi
(B.S., University of Tehran, 2012)
(M.S., The State University of New York at Buffalo, 2014)
May 19, 2018
1
© Copyright by
Saeed Dehghan Manshadi
2018
1
TO MY LOVELY WIFE FOR HER UNCONDITIONAL LOVE AND SUPPORT
AND
TO MY DEAREST PARENTS FOR THEIR LIFELONG LOVE AND ENCOURAGEMENT
vi
ACKNOWLEDGMENTS
I would like to acknowledge the guidance of my advisor, Professor Mohammad Khodayar, not
only for his continuous support in guiding me through my research and completing this dissertation, but
also he has been my mentor any time I needed advice. This dissertation would have not been possible
without his encouragement, guidance, and insight and valuable suggestions. I also appreciate the valuable
comments and suggestions of my advisory committee members, Professors Jianhui Wang, Behrouz Peikari,
Khaled Abdelghany, and Eli Olinick.
I was fortunate enough to work alongside a number of wonderful friends and distinguished
current and former students. I’d like to thank them for their invaluable help on many occasions, spirited
discussions about the research, and more importantly for their friendship and support.
As always, my greatest gratitude and appreciation is to my wife, Parya, my parents, Zinat-Sadat
and Hamidreza for their everyday inspiration and encouragement and endless sacrifices without which I
could never stand here. My sister, Zohreh has always been there for me and I cannot thank her enough for
that.
vii
Dehghan Manshadi, Saeed B.S., University of Tehran, 2012
M.S., The State University of New York at Buffalo, 2014
Microgrids: Resilience, Reliability, and Market Structure
Advisor: Mohammad Khodayar
Doctor of Philosophy degree conferred May 19, 2018
Dissertation completed April 3, 2018
This dissertation addresses several challenges corresponding to the operation and planning of
microgrids in distribution networks including the formation, energy supply resilience, the contribution of
microgrids in the short-term and mid-term operation of bulk power networks and the computational
challenges in operation of interdependent infrastructure systems.
An approach to transforming the active distribution network with distributed energy resources
into multiple autonomous microgrids is presented in the first section. The distribution network consists
of several generation resources and demand entities that could be clustered into autonomous microgrids.
Eigen decomposition in the graph spectra of the distribution network is leveraged to determine the
boundaries for microgrids and a mixed-integer programming problem that minimizes the expansion cost
within microgrids. Once microgrids are formed, in the next section, a framework is proposed to identify
the vulnerable components, and ensure the resilient operation of coordinated electricity and natural gas
infrastructures considering deliberate disruptions. The microgrid demands, which consist of electricity
and heat demands, are served by the interdependent electricity and natural gas supplies. The proposed
approach addressed the vulnerability of multiple energy carrier microgrids to deliberate disruptions, in
order to apply preventive reinforcements to improve the resilience of energy supply. The proposed
methodology is formulated as a bi-level optimization problem to address the optimal and secure
operation of multiple energy carrier microgrids.
To investigate the contribution of microgrids in the bulk power network, in the next section, a
hierarchical structure for the electricity market is proposed to facilitate the coordination of energy
markets in distribution and transmission networks. The proposed market structure facilitates the
viii
integration of microgrids in the electricity markets to provide energy and ancillary services. In the
proposed hierarchical structure, microgrids participate in the energy market at the distribution networks
settled by the distribution network operator, and load aggregators interact with microgrids and
generation companies to import/export energy to/from the distribution network electricity markets
from/to the wholesale electricity market. Furthermore, the impact of microgrids in the mid-term
operation of bulk power systems is investigated. A new framework for risk-averse mid-term generation
maintenance scheduling in the power systems is presented. Microgrid aggregators facilitate the
participation of microgrids in the wholesale market. The effect of microgrids as controllable demand
entities on the generation maintenance scheduling practices in the power system is investigated. The
uncertainties in the marginal cost of generation in microgrids, the generation capacity installed within
the microgrids, and the system electricity demand are captured using respective nominal values and
uncertainty intervals. Moreover, the contingencies in transmission network are addressed by introducing
additional variables. A two-stage robust optimization problem is formulated to determine a trade-off
between the performance and conservativeness of the procured solution in the long-term operation
horizon.
The last section of this dissertation addresses the coordinated operation of interdependent
electricity and natural gas infrastructure systems to improve the security and reliability measures in both
infrastructure systems and mitigate the risk of demand curtailment. The electricity and natural gas
network operation problems are non-convex mixed-integer nonlinear programming problems that are
hard to solve in polynomial time. The non-convex feasible region is formed by the Weymouth constraint
and the introduced binary commitment decision variables in the natural gas and electricity network
operation problems respectively. A sparse semidefinite programming (SDP) relaxation is utilized to
procure the optimal solution for the coordinated operation of electricity and natural gas networks. The
presented algorithm leverages the sparseness of the natural gas network to construct several small
matrices of lifting variables that are used to form a tight and traceable SDP relaxation. A set of valid
constraints that tighten the relaxation ensures the exactness of the solution procured from the relaxed
problem.
viii
TABLE OF CONTENTS
ACKNOWLEDGMENTS ..............................................................................................................................vi
TABLE OF CONTENTS ............................................................................................................................ viii
LIST OF FIGURES ........................................................................................................................................xi
LIST OF TABLES ...................................................................................................................................... xiii
LIST OF SYMBOLS .................................................................................................................................... xv
1 INTRODUCTION ........................................................................................................................................ 2
1.1 Expansion of Autonomous Microgrids in Active Distribution Networks ........................................... 2
1.2 Resilient Operation of Multiple Energy Carrier Microgrids ............................................................... 4
1.3 A Hierarchical Electricity Market Structure for the Smart Grid Paradigm ......................................... 6
1.4 Risk Averse Generation Maintenance Scheduling with Microgrid Aggregators .............................. 10
1.5 Coordinated Operation of Electricity and Natural Gas Systems: A Convex Relaxation Approach .. 12
2 EXPANSION OF AUTONOMOUS MICROGRIDS IN ACTIVE DISTRIBUTION NETWORKS ....... 14
2.1 Graph spectra and Fiedler theory ...................................................................................................... 14
2.2 Problem Formulation ........................................................................................................................ 16
2.2.1 Network topology problem (Upper-level problem) .................................................... 16
2.2.2 Reliability evaluation problem (Lower-level problem) .............................................. 18
2.2.3 Microgrid topology validation problem ..................................................................... 20
2.2.4 Solution Methodology ................................................................................................ 21
2.3 Case Study ........................................................................................................................................ 24
ix
2.3.1 A Sample Meshed Distribution System ...................................................................... 24
2.3.2 The Modified IEEE 123-bus system .......................................................................... 30
2.4 Summary ........................................................................................................................................... 34
3 RESILIENT OPERATION OF MULTIPLE ENERGY CARRIER MICROGRIDS ................................ 35
3.1 Energy Hub ....................................................................................................................................... 36
3.2 Preventive Reinforcement in Cyber Infrastructure ........................................................................... 37
3.3 Problem Formulation ........................................................................................................................ 39
3.4 Case Study ........................................................................................................................................ 45
3.4.1 Case 1: No preventive reinforcement in the microgrid ............................................... 48
3.4.2 Case 2: First Stage of Preventive Reinforcement ....................................................... 49
3.4.3 Case 3: Second Stage of Preventive Reinforcement ................................................... 51
3.4.4 Third Stage of Preventive Reinforcement .................................................................. 52
3.4.5 Preventive reinforcement procedure ........................................................................... 52
3.5 Summary ........................................................................................................................................... 55
4 A HIERARCHICAL ELECTRICITY MARKET STRUCTURE FOR THE SMART GRID PARADIGM
....................................................................................................................................................................... 57
4.1 Problem Formulation ........................................................................................................................ 57
4.1.1 GENCO’s objective .................................................................................................... 58
4.1.2 Microgrid’s objective ................................................................................................. 58
4.1.3 LA’s objective ............................................................................................................ 59
4.1.4 The WEM clearing process ........................................................................................ 60
x
4.1.5 The DNEM clearing process ...................................................................................... 60
4.2 Multi-level complete information game ........................................................................................... 61
4.3 Case study ......................................................................................................................................... 68
4.3.1 Case 1 ......................................................................................................................... 69
4.3.2 Case 2 ......................................................................................................................... 72
4.3.3 Case 3 ......................................................................................................................... 75
4.4 Summary ........................................................................................................................................... 77
5 RISK AVERSE GENERATION MAINTENANCE SCHEDULING WITH MICROGRID
AGGREGATORS ......................................................................................................................................... 79
5.1 Problem Formulation ........................................................................................................................ 80
5.2 Solution Methodology ...................................................................................................................... 84
5.3 Case Study ........................................................................................................................................ 88
5.3.1 6-Bus Power System .................................................................................................. 88
5.3.2 IEEE 118-bus system ................................................................................................. 94
5.4 Summary ........................................................................................................................................... 97
6 COORDINATED OPERATION OF ELECTRICITY AND NATURAL GAS SYSTEMS: A CONVEX
RELAXATION APPROACH ....................................................................................................................... 99
6.1 Problem Formulation ........................................................................................................................ 99
6.1.1 Short-term Operation of Natural Gas Network .......................................................... 99
6.1.2 Short-term Operation of Electricity Network ........................................................... 101
6.2 Proposed Convex Relaxation .......................................................................................................... 102
xi
6.2.1 Background .............................................................................................................. 102
6.2.2 Convex Relaxation of Natural Gas Network Operation Problem ............................. 103
6.2.3 Convex Relaxation of the Electricity Network Operation Problem ......................... 109
6.2.4 Coordination among Electricity and Natural Gas Networks Operation ................... 111
6.3 Case Study ...................................................................................................................................... 112
6.3.1 Scenario 1 – Uncoordinated operation of the electricity and natural gas networks .. 114
6.3.2 Scenario 2 – Coordinated operation of the electricity and natural gas networks ...... 115
6.3.3 Scenario 3 – Scenario 2 with Congestion in the Electricity Network ....................... 117
6.3.4 Scenario 4 – Scenario 3 with Congestion in the Natural Gas Network .................... 118
6.3.5 Discussion ................................................................................................................ 119
6.3.6 IEEE 118-Bus System with 12-junction Natural Gas network ................................. 122
6.4 Summary ......................................................................................................................................... 125
7 SUMMARY ............................................................................................................................................. 126
8 APPENDIX .............................................................................................................................................. 128
BIBLIOGRAPHY ....................................................................................................................................... 130
xi
LIST OF FIGURES
Figure Page
1-1 The hierarchical electricity market structure. ............................................................................................ 8
2-1 The sample connected graph (a). The sample sectioned graph (b). ........................................................ 15
2-2 The spectra of the sample graph. ............................................................................................................ 16
2-3 The proposed electricity distribution network......................................................................................... 27
2-4 Electricity distribution network Case 1 ................................................................................................... 29
2-5 The spectra of the graph associated with distribution network ............................................................... 29
2-6 Electricity distribution network in Case 2 ............................................................................................... 30
2-7 Modified IEEE 123-bus test system ........................................................................................................ 34
3-1 Energy hub with electricity, natural gas and heat ................................................................................... 36
3-2 Disruption of service in multiple energy carrier microgrids ................................................................... 38
3-3 Hyper plane for the Weymouth equation ................................................................................................ 45
3-4 Multiple energy carrier microgrid. .......................................................................................................... 46
3-5 Microgrid operation condition in Case 1 ................................................................................................. 49
3-6 Microgrid operation condition in Case 2 ................................................................................................. 50
3-7 Microgrid operation condition in Case 3 ................................................................................................. 52
3-8 Operation and reinforcement costs of multi-energy carrier microgrid. ................................................... 55
4-1 Competition in the hierarchical electricity market .................................................................................. 66
4-2 Power network topology ......................................................................................................................... 69
4-3 Bidding strategies of market participants, exchanged power, and energy price in Case ......................... 71
4-4 Market clearing results in Case 1 ............................................................................................................ 72
xii
4-5 Market clearing result in Case 2.1 ........................................................................................................... 73
4-6 Market clearing results in Case 2.2 ......................................................................................................... 74
4-7 Market clearing results in Case 3 ............................................................................................................ 76
4-8 Bidding strategies of market participants and energy prices in Case 3 ................................................... 77
5-1 Demand profile for one year ................................................................................................................... 89
5-2 Scheduled capacity on outage in Case 1 ................................................................................................. 90
5-3 Scheduled capacity on maintenance in Case 1 with the increase in the transmission line capacity ........ 90
5-4 Demand profile with and without microgrids ......................................................................................... 91
5-5 Increase in the total operation and maintenance cost because of the increase in the budget of uncertainty
....................................................................................................................................................................... 92
5-6 Demand profile for IEEE 118-bus power system ................................................................................... 94
5-7 Marginal cost of the local generation for the aggregated microgrids ...................................................... 95
5-8 Sensitivity of the total operation and maintenance cost to the installed generation capacity and marginal
cost multiplier (MCM) of microgrids ............................................................................................................ 96
6-1 The algorithm to procure an optimal solution for the natural gas network operation problem. ............ 104
6-2 The 6 bus-system interconnected with 7-junction natural gas .............................................................. 113
6-3 The normalized demand profile for electricity and natural gas ............................................................. 113
6-4 The generation dispatch of G1 in scenarios .......................................................................................... 121
6-5 The generation dispatch of G2 in scenarios .......................................................................................... 121
6-6 The generation dispatch of T1 in scenarios ........................................................................................... 121
6-7 The flow of natural gas in pipeline PL3 in scenarios 3-4. ..................................................................... 122
6-8 The profile for the natural gas volume withdrawn from sources in scenarios 3-4. ............................... 122
6-9 12-junction natural gas network ............................................................................................................ 123
xiii
LIST OF TABLES
Table Page
2-1 Power generation unit characteristics ...................................................................................................... 25
2-2 Required LOEP at each bus .................................................................................................................... 26
2-3 Distribution line characteristics ............................................................................................................... 26
2-4 The LOEP in distribution network with no DER and microgrid ............................................................. 26
2-5 The LOEP of demand at each bus in case 1 ............................................................................................ 28
2-6 The LOEP of demand at each bus in Case 2 ........................................................................................... 30
2-7 Required LOEP of demands in IEEE-123 bus system ............................................................................ 33
3-1 Natural gas power generation unit characteristics ................................................................................... 46
3-2 Electricity and heat demand .................................................................................................................... 46
3-3 Distribution network characteristics (electricity and natural gas) ........................................................... 47
3-4 Outcomes of preventive reinforcement procedure .................................................................................. 54
4-1 Transmission and distribution network data ........................................................................................... 68
4-2 Generation cost curve of microgrids and GENCOs ................................................................................ 68
4-3 Bidding strategy and awarded dispatch in Case 1 ................................................................................... 72
4-4 Bidding strategy and awarded dispatch in case 2 (L1 is congested) ....................................................... 73
4-5 Bidding strategy and awarded dispatch in case 2 (L4 is congested) ....................................................... 74
4-6 Bidding strategy and awarded dispatch in case 3 .................................................................................... 75
5-1 Transmission line characteristics ............................................................................................................ 89
5-2 Thermal units characteristics................................................................................................................... 89
xiv
5-3 Scheduled maintenance in weeks for cases 1-4....................................................................................... 96
5-4 The periods of transmission network contingencies for cases 1-4 .......................................................... 96
6-1 Generation unit charcteristics ................................................................................................................ 114
6-2 Transmission line characteristics .......................................................................................................... 114
6-3 Natural gas resources characteristics..................................................................................................... 114
6-4 Natural gas pipeline characteristics ....................................................................................................... 114
6-5 Hourly unit commitment in Scenario 1 ................................................................................................. 115
6-6 Natural gas pressures in scenario 1 at hour 17 ...................................................................................... 115
6-7 Hourly unit commitment in Scenario 2 ................................................................................................. 116
6-8 Natural gas pressures in scenario 2 at hour 17 ...................................................................................... 116
6-9 The ratio of the two largest eigenvalues in scenario 2 at hour 17 ......................................................... 117
6-10 Natural gas pressures in scenario 3 at hour 17 .................................................................................... 117
6-11 Natural gas pressures in scenario 3 at hour 17 .................................................................................... 118
6-12 The ratio of the two largest eigenvalues in scenario 3 at hour 17 ....................................................... 118
6-13 Natural gas pressures in scenario 4 at hour 17 .................................................................................... 119
6-14 The ratio of the two largest eigenvalues in scenario 4 at hour 17 ....................................................... 119
6-15 Hourly unit commitment in Scenario 1 ............................................................................................... 124
6-16 Changes in the hourly commitment in scenario 2 from scenario 1 ..................................................... 125
xv
LIST OF SYMBOLS
sjA Natural gas junction-resource incidence matrix
bB Set of units that are connected to bus b
(.),(.)B Imaginary part of microgrid admittance matrix
,o oj lB B Set of buses o corresponds to the market participants j and l
,j ob Susceptance of the branch between energy hub j and o
pC Pipeline constant
(.) _(.)C Coupling factors in energy conversion matrix
,t sgc Price of electricity at time t in scenario s
,,
g mb t
C Uncertain variable of marginal cost of segment m of aggregated microgrid on bus b at
time t
,0,
gb t
C Marginal cost of the microgrid on bus b at time t
siC Marginal cost for segment s of the cost curve for unit i
E Budget of uncertainty
bEENS Expected energy not supplied at bus b
,b bF F Minimum/maximum limits for power flow in branch b
,c iF Production cost function for thermal unit i
xvi
c,sF Natural gas supply volume cost at source s
,pj of Natural gas flow in pipeline p between energy hubs j and o
,l tf Power flow of the transmission line l at time t
(.),(.)G
Real part of microgrid admittance matrix
jGG Set of natural gas-fired units connected to junction j
bGG Set of units that are connected to bus b
jGS Set of natural gas resources connected to junction j
bGR Set of the utility grid connection to bus b
, ,,j o j og y
Conductance and admittance of the distribution line between energy hub j and o
(.)iH Natural gas consumption function for natural gas generation unit i
jHH Set of non-electric natural gas demands connected to junction j
,i tHF The number of periods in which the unit i was on maintenance at the beginning of period
t
,i tHO The number of periods in which the unit i was available at the beginning of period t
I Identity matrix
,i tI Commitment decision of unit i at hour t ; 1 if the unit is ON and 0 otherwise
k The number of contingencies in each period
bLOEP Loss of energy probability at bus b
xvii
, ,,f j t jL L Set of distribution lines starting from /ending at energy hub j
L Total number of considered contingencies in the operation horizon
(.)M Required resources for interdiction of each component type in microgrid
M Total available resources for interdiction
iMC Maintenance cost of unit i
iMHO Maximum number of periods in which unit i is available
iNHO Minimum number of periods in which unit i is available
(.),d
jP Served demand in energy hub j
h
kP Heat energy produced by heater k
,Qe ei iP Real/reactive generation dispatch of unit i
,inj injj jP Q Real/reactive power injection at energy hub j
,i djP Natural gas consumption of units i connected to energy hub j
,, ,j o j oPL QL Real/reactive power flow between energy hubs j and o
, .,,
,Qd s d sb tb t
P Real/reactive demand served at bus b at time t in scenario s
, ,,Qt s t si iP Real/reactive generation dispatch of unit i at time t in scenario s
, ,,Qt s t sg gP Real/reactive power dispatch of main grid g at time t in scenario s
xviii
, ,, ,
,inj s inj s
b t b tP Q Real/reactive power injection at bus b at time t in scenario s
, ,,
, ,t s t sb b b b
PL QL Real/reactive power flow between buses b and b at time t in scenario s
max max,Qi iP Real/reactive generation capacity of unit i
, ,, ,
,QD s D sb t b t
P Real/reactive power demand at bus b at time t in scenario s
(.), ,,D e D
j jP Q Active/reactive electricity demand at energy hub j
,f jP Set of pipelines starting from energy hub j
,t jP Set of pipelines ending at energy hub j
max max,Qi iP Maximum real/reactive generation of unit i
jP
Generation dispatch of generator j [MW]
mP
Generation dispatch of microgrid m [MW]
exp imp,
l lP P
The power “exported from” / “imported to” DNEM by LA l [MW]
DP Total demand in the distribution network [MW]
DoP The demand on bus o in the WEM [MW]
exp exp,
l lP P
Minimum/maximum power “exported from” the DNEM by LA l [MW]
imp,
impl l
P P
Minimum/maximum power “imported to” the DNEM by LA l [MW]
(.) (.),P P
Minimum/maximum generation dispatch [MW]
xix
,,g m
b tP Generation of segment m of aggregated microgrid connected to bus b at period t
,i tP Generation dispatch of unit i at period t
,d
b tP Demand on bus b at time t
,0,d
b tP Demand on bus b at time t
,
h
k tP Non-electric natural gas demand k at hour t
,D
b tP Electricity demand in bus b at hour t
,f jP Set of pipelines starting from junction j
,t jP Set of pipelines ending at junction j
,Qe dj Served reactive electricity demand of energy hub j
jR Payoff function of GENCO j [$]
mR Payoff function of microgrid m [$]
,l lR R Payoff function of LA l in the DNEM/WEM [$]
iRHF Required number of periods for maintenance of unit i
iRU Maximum ramping up of unit i
iRD Maximum ramping down of unit i
,b br Resistance of the line between buses b and b
xx
,,
t sb b
SL
Apparent power flow between buses b and b at time t in scenario s
max,b bSL Maximum capacity of line connecting buses b and b
blSF Shift factor of line l with respective to bus b
o
bSF Shift factor of branch b with respect to injection at bus o
,l bS Element of transmission line-bus incidence matrix
,offon
i iT T Minimum up/down time of unit i
,b bu Decision variable for the connectivity between buses b and b
, ,,d db t b tu v Binary variables representing the uncertainties in demand
, ,,b t b tu v Binary variables representing the uncertainties in generation capacity of microgrid
, ,,
g gb t b t
u v Binary variables representing the uncertainties in marginal cost of microgrid
,l tu Binary variable representing the uncertainties in availability of components
,t siUX Generator’s outage status at time t in scenario s, 1 if available, otherwise 0
iUX Availability status of generator unit i , 1 if interdicted, otherwise 0
,,
t sb b
UY Distribution line’s outage status at time t in scenario s, 1 if available, otherwise 0
,lj oUY Availability status of transmission line l between energy hubs j and o , 1 if interdicted,
otherwise 0
,t sgUZ The main grid connection's outage status at time t in scenario s, 1 if available, otherwise 0
pUZ Availability status of natural gas pipeline p , 1 if interdicted, otherwise 0
xxi
max min,s sv v Max/min natural gas supply volume of resource s
sv Natural gas supply volume of resource s
,t sb
V Voltage magnitude at bus b at time t in scenario s
max minV ,V Maximum and minimum voltage magnitude
jV Voltage magnitude at energy hub j
(.)VOLL Value of lost load
, ,,j o j ox r Line reactance and resistance between energy hubs j and o
lX Inductive reactance of the transmission line l
,i tX Binary variable representing the transition to outage stage for maintenance of unit i at
period t
,oni tX On time of unit i at hour t
,offi tX Off time of unit i at hour t
,b by Admittance of the line between buses b and b
,i tY Binary variable representing the transition from outage to available mode for unit i at
period t
, ,i i i Fuel cost coefficients of generation unit i
,t sb
Voltage angle at bus b at time t in scenario s
xxii
s Probability of scenario s
t Duration of time period in load duration curve
.b b Auxiliary parameter
j Natural gas pressure at energy hub j
j Voltage angle at energy hub j
max min, Max/min natural gas pressure at each energy hub
max min, Max/min voltage angle at each energy hub
j Initial natural gas pressure at energy hub j
m lρ ,ρ Bidding vector of microgrid m/ LA l in the DNEM [$/MW]
,j l ρ ρ Bidding vector of GENCO j/ LA l in the WEM [$/MW]
m lψ ,ψ
Bidding strategy vector of microgrid m/LA l in the DNEM
j l ψ ,ψ
Bidding strategy vector of unit j of GENCO i/LA l in the WEM
τ ,φ Vector of slack variables
, , , θ π θ π Vector of Lagrangian multipliers
0 0, , ,b b Lagrange multipliers
(.) (.)ψ ,ψ
Upper/lower limit of the bidding strategy vector in the DNEM
xxiii
(.) (.)ψ ,ψ Upper/lower limit of the bidding strategy vector in the WEM
,b t Uncertain variable for generation capacity of microgrid on bus b at time t
(.) (.)(.) (.)
, Lagrangian Multipliers
0,b t Generation capacity of microgrid on bus b at time t
Convergence tolerance
,gb t
C Deviation in marginal cost of the microgrid on bus b at time t
,d
b tP Deviation of demand on bus b at time t
,b t Deviation of generation capacity of microgrid on bus b at time t
,pj o Maximum compression ratio for natural gas pipeline p
s Cost of natural gas supply at source s
, Minimum/Maximum natural gas pressure
,s sv v Max/min natural gas withdrawal of resource s
2
Chapter 1
1INTRODUCTION
1.1 Expansion of Autonomous Microgrids in Active Distribution Networks
Reliability of energy supply in distribution networks is dependent on the availability of the distribution
feeders, as they have radial topology [Bar89]. A fault or failure in the line or feeder within the radial
network will lead to demand curtailment. These failures are caused randomly by component outages in the
network [DOE10]. The increase in the installed capacity of DERs requires distributed and efficient control
of such resources, which is facilitated by forming microgrids [Nak13]. Microgrids are composed of DERs
and demands with distinct boundaries that are connected to the utility grid through the point of common
coupling (PCC) or operated in island mode [Iag13]. Forming microgrids can help to improve the restoration
capability of the distribution networks [Li14], [Gao16]. Islanding in microgrids reduces the adverse effects
of outages and power quality deterioration on local demands. Moreover, developing efficient control
schemes in microgrids facilitates the trade of electricity and ancillary services in distribution market
[Man16].
While increasing the number of microgrids promotes the service reliability, facilitates distributed control
over the generation and demand resources and reduces the investment cost to improve the reliability and
resilience; defining the boundaries for microgrids in distribution networks to serve the customers is a
challenging task. In this context, partitioning the electricity network into several sub-networks using
graphical and Eigenvalue sensitivity-based approaches is discussed in [Roy00]. The graph spectra are
assessed using a sensitivity-based approach in order to partition the network and analyze the dynamic
properties of the power system in the procured sub-networks. However, the base for network partitioning
does not capture the power flow constraints and the reliability of the power system.
While earlier research is focused on maintaining the supply adequacy in contingencies, chapter addresses
the supply adequacy capturing the reliability requirements of the demands and the quality of service at
demand buses is ensured by forming microgrids and expansion of the distribution network. As a
3
consequence, the developed microgrids provide heterogeneous energy supply reliability at demand buses
considering the outages in generation and distribution assets. Sectionalizing the distribution network to
provide boundaries for the DERs to serve the consumers will increase the reliability of energy supply in
contingencies as DERs are dispatched as backup generation resources in the distribution network.
Moreover, sectionalizing the distribution network to form microgrids will improve the robustness in the
control of generation and demand assets. In this context, DERs operate within the microgrids to ensure the
reliability of energy supply. A fundamental challenge in controlling DERs is the significant number of
control variables in the distribution network [Las11]. When DERs serve the customers in contingencies, a
failure in a complex control system can shut down the entire system considering the fact that the generation
resources and their inertia are limited in the distribution network. In order to mitigate the disturbance
propagation and reduce the vulnerability of the distribution networks to voltage sags, swells, faults, and the
uncertainties in demand and supply balance, microgrids are formed and islanded in emergency conditions.
Islanding improves the quality of energy supply during disturbances in the network. Minimizing the
number of generation, distribution and demand assets in each microgrid – to provide more robust control –
will maximize the number of microgrids formed in the network. However, the reliability requirement of the
demand will limit the size of the microgrids. Larger microgrids with more DERs provide higher service
reliability. Moreover, expanding the distribution network within a microgrid improves the service
reliability. Therefore, the size of the microgrids is determined by a trade-off between the controllability of
microgrids and the reliability of service and network expansion within microgrids improve the reliability
further.
Service restoration and reconfiguration performed by distribution management systems ensure the
reliability of energy supply for the consumers; however, managing large-scale distribution automation
systems with large number of DERs, switches, and controllable demands is a challenging task. Developing
a robust real-time energy management framework with a large number of control variables is
overwhelming for a central controller in the distribution network. Moreover, distributed energy
management solutions to control the energy flow between the interconnected microgrids may not provide
robust and reliable solutions [Saa12]. Hence, sectionalizing the distribution networks, and islanding the
4
microgrids will reduce the control processing burden and mitigates the disturbance and uncertainty
propagation.
Allocating the generated energy to demand buses in each microgrid based on the ownership of the
generation assets could be another criterion to determine the PCCs in the distribution networks. DER owner
may offer to sell electricity to nearby demand buses within the microgrid [Rah07]. The distribution network
operator or distribution company may own the DERs in the system [Imi15]-[Act15]. In such cases, the
presented formulation could be used to assign certain consumers to the DERs and form islanded microgrids
in contingencies within the distribution network.
1.2 Resilient Operation of Multiple Energy Carrier Microgrids
The increase in penetration level of distributed energy resources (DERs) supplied by natural gas in
electricity networks warrants the coordinated operation of electricity and natural gas distribution networks.
The rising concerns over the reliability and quality of service in energy distribution networks promotes the
concept of multiple energy carrier microgrids. Multiple energy carrier microgrid is referred to as an
interconnected electricity and natural gas distribution network in which the electrical network is composed
of a group of interconnected electrical demands and DERs that represent a single controllable entity within
the electricity grid with the ability to operate in grid-connected and island modes. The natural gas
distribution network within the microgrid is composed of several source points which are connected to the
gas distribution network and a group of load points which represent electrical generation units or heat
demands. Several literatures have addressed the interdependence of electricity and natural gas
infrastructures in power systems. The model in [Que07] presents a fundamental understanding of the
interdependent energy systems including coal, natural gas and electricity while ignoring the network
constraints. The interdependence of natural gas and electricity infrastructures is discussed in [Sha05],
where the optimal operation of the interconnected natural gas-electricity network is affected by several
factors including, physical characteristics of natural gas pipeline and electricity networks, operational
procedures in electrical and natural gas distribution networks, types of electricity generating plants,
availability of natural gas and electricity supply, transmission and delivery constraints in natural gas and
electricity networks, and their respective volatile market prices. In [Li08], switching between fuel supplies
5
in electricity generation is proposed as an effective approach to perform peak shaving on natural gas
demand while enhancing the power system operation security. Several factors including the natural gas
market price, natural gas pipeline pressure loss, and simultaneous demand peak in electricity and natural
gas which causes price spikes in severe weather conditions and possible outages in natural gas pipelines;
affect the power system operation by increasing congestions, price of electricity or even demand
curtailment in severe conditions. A security-constrained short term generation scheduling in electrical
networks is developed in [Liu09]. The slow transient process in natural gas transmission networks was
considered by a set of partial differential and algebraic equations in [Liu11]. The proposed approaches
addressed the effects of natural gas transmission network on power system security by implementing
determined scenarios for possible contingencies in both energy infrastructures.
A framework for the comprehensive modeling of multiple energy delivery systems in which energy is
converted within an energy hub is presented in [Kra11]. Energy hub provides the features of input, output,
storage, and conversion of multiple energy carriers. An integrated optimization framework which
incorporates a simplified model for the network flow of the interdependent energy carriers is proposed in
[Gei07]. An approach for optimal scheduling of multiple energy networks including electricity, natural gas,
and district heat is introduced in [Gei06]. A linear formulation is proposed in [Alm11] to solve the large-
scale optimal energy flow in multiple energy delivery networks. An expansion planning framework within
the electricity and natural gas distribution networks with high penetration of gas-fired distributed
generation is presented in [Sal13], which incorporates heuristic methodologies to find the optimal
expansion planning strategies. Discussion in [Uns10] expressed the savings gained by optimal operation of
integrated electricity and natural gas networks. In [Awa09], an integrated formulation of electricity and
heat distribution networks to procure the optimal electricity and heat flow within a microgrid is presented.
Chapter 3 addressed the coupling between electricity and natural gas energy infrastructures within the
microgrids which can affect the resilience of energy supply. The resilience of energy infrastructure is
addressed by identifying vulnerable components in electricity and natural gas infrastructures. Deliberate
interruptions do not occur frequently, but, when they do, they can be tragic. Quantitative probabilistic
evaluation of such low-probability-high-consequence contingencies is very challenging, resource-
6
demanding, and subjected to inaccuracies. Moreover, the outcome of a disruption strongly depends on the
system operation conditions prior to the attack, as well as the available resources to the offenders, both of
which are also highly uncertain. There are a number of publications addressed the contingencies as a
consequence of threat in power systems [Tep12],[Sal04],[Bie07][Rom12],[Yao07]. Contingencies as a
result of natural disasters, physical or cyber-attacks would extinguish the key elements of the electricity
infrastructure and may lead to blackouts. The disruptions of energy supply in electric power grids caused
by physical attacks are addressed in [Sal04] which is presented as a bi-level mixed integer programming
problem and solved by a heuristic approach. A model to protect the power network against a range of
scenarios which lead to disruption in energy supply and blackouts is developed in [Bie07]. The three-level
attacker-defender-planner model presented in [Rom12] as a multi-level mixed integer programming
problem finds the optimal defense strategy in power system. Here, the interdependence between energy
networks was pointed out as a potential improvement in finding the more realistic defense strategies. In
[Yao07], a tri-level optimization model is proposed for power network defense which is solved iteratively
by forming nested bi-level optimization problem with budget constraints associated with defender and
attacker assuming that the defended components are infallible once they are hardened. The physical
disruption, modeled as a mixed-integer bi-level programming (MIBLP) problem in [Mot05] ignores the
interdependence of electricity and natural gas infrastructures. Here, the power system under attack is
represented by dc power flow approximation, without offering any preventive guidelines for reinforcing the
system. Other publications emphasized the importance of considering the interdependencies between
natural gas and electricity to improve resilience and reliability measures without quantifying the outcomes
[Zim09],[Rin04]. Since microgrids are developed to provide higher power quality and reliability for
consumers, the resilience of energy supply is an important concern within microgrids.
1.3 A Hierarchical Electricity Market Structure for the Smart Grid Paradigm
Developing the hierarchal market structure composed of generation entities and demand aggregators
facilitates the participation of small demand entities and end customers in the electricity market. With the
advent of active distribution networks with distributed energy resources, responsive demands, and
intelligent controls, there is a concern on the increased number of intelligent supply and demand entities in
7
the distribution network that affects the wholesale electricity market operation. Scheduling of such entities
by ISO will increase the scale of the day-ahead and real-time scheduling problem and corresponding
computational burden. Moreover, the competition among these entities are not fully addressed as a result
technical limitations and lack of effective business model in this paradigm. Chapter 4 proposed a
framework to facilitate competition among larger number of market participants in the WEM and the
DNEM. The presented framework is further extendable to support the participation of microgrids in the
WEMs. LAs were considered as intermediate agents between the utility operator and consumers to perform
Demand Response (DR) in [Gka13],[Par13],[Zug13]. While the proposed structures addressed the
interaction between the customers and aggregators who respond to the price of electricity, the pricing
scheme does not reflect the synergy between the utility grid, aggregator, and the customers. As microgrids
are deployed in low or medium voltage distribution networks, they exchange energy with the main grid
through an aggregator which represents the middle agent who interacts with the microgrid and the WEM.
Distribution Companies (DISCOs) participate in the WEM and the DNEM and interact with GENCOs,
distributed generations (DG) and interruptible loads (IL). DISCOs bid for energy and ancillary services in
the WEMs [Mas11]. DGs and ILs controlled by DISCOs will benefit from participating in electricity
markets while DISCOs are considered as agents with no financial benefits in [Hag12], [Doo12]. A multi-
period framework for participating DISCO in day-ahead electricity market is proposed in [Li07] using bi-
level optimization technique to maximize the DISCO’s profit while minimizing the operation cost of the
system. A framework to facilitate the participation of DISCOs in a day-ahead bi-lateral and pool energy
markets is presented in [Pal05]. In these publications, the interaction between DGs, ILs and DISCOs
through competition was not addressed. In [Akb11] the synergy between DISCOs and GENCOs is
considered in the WEM where DISCOs bid based on the available capacities of DGs and ILs, and GENCOs
bid based on the marginal cost of the generators owned. DISCO can also participate in the DNEM operated
by the DNO [Lop11]. Here the interaction between DISCOs and GENCOs in the WEM is not considered.
The synergy between DERs and demand resources within the microgrids with the energy service providers
(ESP), who participate in the energy market is addressed in [Asi13], while the interaction between the
WEM and local DNEM is ignored. Several research efforts presented frameworks for the participation of
microgrids and DISCOs in a competitive electricity market by a two-stage hierarchical optimization
8
framework for the day-ahead and real-time electricity markets [Kim11],[Alg09],[Saf13]. However, the
impact of the microgrids on the locational marginal prices (LMPs) of electricity in the WEM was not
addressed.
The synergy between the microgrids and the WEM to maximize the social welfare is presented in
[Zha14]. The electricity market is represented by randomize auction framework ignoring the role of
“microgrid aggregators” as a broker with respective bidding strategies. Direct participation of microgrids in
the WEM is not applicable for the following reasons: 1) technical limitations on voltage levels, generation
and demand capacity, 2) lack of an efficient business model and technical infrastructure to provide
microgrids with access to Open Access Same-Time Information System (OASIS) and Open Access Non-
discriminatory Transmission Services, 3) shortage in an efficient framework to evaluate the potential
services provided by microgrids in the WEM, and 4) increase in the number of microgrids, with diverse
capacities and coverage areas; and the technical restraints on transmitting electricity to the bulk power
network including congestion and voltage constraints [Sha12], [Koe04].
A system of system (SoS) framework is proposed in [Kar14] to address the interaction between the
Figure 1-1 The hierarchical electricity market structure.
Wholesale Electricity Market (WEM)
GENCO j
LA l
MG m
Distribution Network Electricity Market (DNEM)
9
DISCOs and microgrids as independent systems, which exchange energy. The proposed approach
addressed the optimal operation of the distribution network and does not consider any interaction between
DGs and DISCOs to maximize the agent’s payoff in the WEM or the DNEM. In the coordinated energy
management structure presented in [Wan15], microgrids and the DNO exchange electricity to minimize
their operation cost, while the DNO exchanges electricity with the WEM. However, the electricity prices
are considered as parameters in the coordinated energy management structure. Hence, no mechanism is
proposed to represent the simultaneous active participation of LAs in the DNEM and the WEM. In chapter
4, LAs are considered as brokers who participate in the DNEM, which is cleared by the DNO as well as the
WEM settled by the ISO. Introducing LAs in the proposed framework reduces the number of market
participants in the WEM and promotes the competition among participants in the WEM and the DNEM.
Here DGs and ILs are merged to form microgrids and the new market paradigm facilitates the participation
of multiple microgrids in the WEM. In the proposed hierarchical market structure, each market participant
will categorize the unknown strategies and information correspond to other market participants by realizing
several “types”, which eventually would transform the incomplete information game into a complete
information game with imperfect information using the joint probability distribution function to address the
uncertainties associate with the “types” of market participants. In this chapter, the complete information
game between the market participants is presented and the incomplete information game is considered as an
extension to the proposed approach. The main contributions of this chapter are summarized as follows: 1)
The synergies between the WEM and the DNEM are presented. In this paradigm, each LA is an
intermediate agent that participates in the WEM and the DNEM. Hence the strategy chosen in one market
would impact the strategy taken in the other. 2) The proposed hierarchical electricity market structure
provides the required infrastructure for microgrids to participate in energy market in the distribution and
transmission networks, without increasing the computational burden on the ISO associated with the
increased number of market participants. 3) The proposed hierarchical market structure is an application of
dynamic game to facilitate the synergy between multiple electricity markets, where the WEM interacts with
the DNEM [Bas95]-[Rom97]. Figure 1 shows the proposed hierarchical market structure that composed of
these markets. Here, each market participant bids in associated electricity market, while each LA is an
intermediate agent that participates in both markets to address the synergy of the two markets. The LA bids
10
in the WEM and presents the awarded dispatch in the DNEM as demand on its bus in the WEM. The ISO
clears the market and provide the awarded dispatch for the LA and LMP at the LA’s bus. The LA bids in
the DNEM and presents the awarded dispatch in the WEM as demand on its bus in the DNEM. The DNO
clears the market and provide the awarded dispatch for the LA and the market clearing price (MCP) of the
market.
1.4 Risk Averse Generation Maintenance Scheduling with Microgrid Aggregators
Generation maintenance scheduling determines the most effective periods for planned generation
outages in order to maintain the reserve capacity margin and avoid costly demand curtailments in the power
systems. Such practices could be coordinated with transmission maintenance scheduling [Fu07]. The long-
term generation and transmission maintenance scheduling is coordinated with short-term generation
scheduling in [Fu09] to improve the security of the power system. The coordinated generation and
transmission maintenance scheduling that captures the degradation of generation and transmission assets
and equipment malfunction because of loading or ambient temperature and weather condition is addressed
in [Wu10]. The uncertainties in the long-term and short-term operation including the random outages of the
generation and transmission units, load forecast errors and fuel price fluctuations are captured by the
proposed stochastic model for the coordinated generation and transmission maintenance scheduling in
power systems in [Wan16a]. In [Ji16] a risk-averse approach for generation maintenance scheduling in
power systems with high penetration of renewable energy resources is proposed. In [Wan16b] the
generation maintenance scheduling is coordinated with transmission maintenance practices considering the
N-1 contingencies in the generation and transmission components. Other factors including renewable
resources and demand response practices were addressed in the generation maintenance scheduling in
[Mol17] and [Wu08a] respectively. While earlier publications were focused on procuring the outage
schedule of the generation and transmission components, the impact of demand-side management and load
control on the generation maintenance scheduling were not addressed. The demand realized by bulk power
systems could be regulated using the controllable generation assets in the distribution networks to provide
improved solutions for maintenance scheduling while maintaining the security and reliability of the bulk
11
power system. In this context, microgrids are among the most viable solutions to regulate the demand in the
power systems. Microgrids are considered as autonomous electric power systems with defined boundaries,
local demand, generation, and/or storage facilities that can operate in grid-connected or island mode [Mic].
Regulating the demand by leveraging local generation assets in microgrids improves the economics and
security measures of the generation maintenance scheduling practices. However, the uncertainty inherently
exists in the available generation capacity in the microgrids, the marginal price of generating electricity and
the demand that should be partly or completely served by bulk power system. Furthermore, the possible
contingencies in transmission network will affect the power flow and the maintenance scheduling of
generation units. Such uncertainties should be captured in the generation maintenance scheduling problem
as they may result in economic inefficiencies and further jeopardize the long-term security of the power
system. The uncertainties in power systems are captured using stochastic programming approach that
leverages the probability distribution of the uncertain variables such as demand, renewable generation, and
availability of the system components [Ozt05],[Wu07],[Wan08],[Rui09],[Tou09][Wan12]. Such approach
provides improved performance over the deterministic solutions, however, its computation burden will
increase with the increase in the number of incorporated scenarios. Furthermore, the probability distribution
functions for the uncertain variables may not be readily available. In order to address these challenges,
robust optimization (RO) is introduced and applied to several engineering problems [Ben09],[ Ber11]. The
solution to RO problems provides risk-averse strategies by capturing the uncertainty boundaries without
requiring the probability distribution function of the uncertain variables. Such solutions are favorable for
long-term operation planning of power systems including maintenance scheduling as the security of the
power system is maintained considering the worst realization of the uncertain variables. Furthermore,
earlier publications [Fu07], [Fu09], [Wan16a], [Wu08b] addressed maintenance as a one-time practice in
the operation horizon, while the minimum and maximum periods among sequential maintenance practices
were ignored. Improving the mathematical model for the maintenance scheduling by addressing such
limitations enables multiple maintenance practices for the generation units in the longer operation horizon.
12
1.5 Coordinated Operation of Electricity and Natural Gas Systems: A Convex Relaxation
Approach
Several research efforts addressed the interdependence among natural gas and electricity networks
[Sha14],[FMO15]. In the electricity network, the natural gas generation units are fast response units that
compensate for the generation scarcity that is triggered by sudden changes in the renewable generation
dispatch or the electricity demand. The essence of capturing natural gas constraints in bulk-power and
distribution network operation is addressed in [Sha05] and [Gei07] respectively. Some research works
ignored the network model for the electricity or natural gas system and addressed the economic and
environmental objectives/constraints by simply balancing the demand and supply using single-node
approach [Mar16]-[Bah16]. Other research works highlighted the challenges associated with the non-
convexity of the network models [Gei07] by presenting a linear representation [Li08], a multi-dimensional
piecewise linear approximation [Sha17], adaptive partitioning [Wu17], or successive linear formulation
[Abd15] to convexify the network constraints and to determine an equilibrium point [Mar12] for the overall
energy flow. These approaches require a large number of piecewise estimates and a considerable number of
iterations for accurate piecewise linear approximation, while there is no guarantee for the exact solution. In
[Liu09], a security-constrained unit commitment problem is solved using Benders decomposition approach
in which the constraints of the natural gas network are incorporated in the sub-problem. The methodology
proposed in [Liu09] to solve the nonlinear and nonconvex natural gas operation sub-problem is an iterative
procedure with a predefined tolerance for error, and Benders decomposition is used to ensure the feasibility
of the natural gas network constraints. Other research efforts leveraged heuristic approaches such as
genetic algorithm [Moe14] to solve for the energy flows in the interdependent electricity and natural gas
networks. Similar convexification, linearization [Uns10], and heuristic approaches [Mar13] were used for
the expansion planning practices in the electricity and natural gas networks. Several scenarios for
coordination among the electricity and natural gas networks are investigated in [Zlo17] to improve the
economic and security measures. An important notion which was ignored in earlier research is that the
electricity and natural gas systems are usually operated by different entities with heterogeneous objectives
[Qad14]. Therefore, the short-term and long-term operational planning for these interdependent
13
infrastructure systems should capture the limitations on the shared information among their respective
system operators [He17]. A consensus operation framework facilitates the coordinated operation of
electricity and natural gas networks. The benefits of the consensus-based distributed approach over the
Lagrange relaxation are presented in [Wen17]. In order to guarantee the convergence, the short-term
operation problem for each infrastructure system is required to be formulated as a convex optimization
problem [He17].
In the natural gas network, the flow in the pipelines is formulated by a nonconvex Weymouth constraint
and in order to convexify this constraint, the direction of natural gas flow in the pipeline is assumed as
known and unchanged [He17]. This assumption may not be valid as the direction of the natural gas flow in
the pipeline may change suddenly to supply the natural gas generation units and compensate for the
variability of renewable generation in electricity networks [Cor14],[Hib12]. The variability in the
generation dispatch of natural gas generation units is referred to the intra-day changes in the dispatch of
these units in response to hour-to-hour changes in system demand that cannot be served by the renewable
generation and other thermal generation technologies. Furthermore, despite knowing the direction of the
natural gas flow, the Weymouth constraint is nonlinear [San16]; therefore, linearization techniques were
used to formulate the electricity network operation problem with natural gas system constraints as a mixed
integer programming (MIP) problem with linear constraints [Li08],[Sha17],[Abd15].
14
Chapter 2
2EXPANSION OF AUTONOMOUS MICROGRIDS IN ACTIVE DISTRIBUTION NETWORKS
The objective of this chapter is to develop a framework for the expansion of the distribution network,
while maximizing the number of autonomous microgrids to maintain the reliability and security of the
energy supply. The procured PCCs determine the boundaries of the microgrids and disjoint the microgrids
from the distribution network in contingencies. The expansion of the distribution lines improves the
reliability of energy supply within microgrids. The main contributions of this chapter are summarized as
follows:
• A bi-level optimization problem is formulated to identify the PCCs by maximizing the number of
microgrids considering the reliability requirements of the customers.
• The expansion of distribution lines ensures the reliability and security of energy supply in the island
operation.
• Eigen decomposition of the Laplacian matrix of the distribution network graph is used to determine the
graph spectra and formulate the connectivity of the distribution network graph.
• The uncertainties in demand and the availability of the components within the microgrids are captured
by developing scenarios using the Monte Carlo simulation.
The rest of this chapter is organized as the follows. In Section 2.1, the application of Fielder theory to
obtain the graph spectra of the distribution network and reconfigure the distribution network to form
microgrids is presented. The problem formulation for the expansion of the distribution network to achieve
the required reliability is discussed in Section 2.2. The effectiveness of the proposed framework is
examined in multiple case studies in Section 2.3. Finally; the summary are presented in Section 2.4.
2.1 Graph spectra and Fiedler theory
In this section, the spectra of the network are utilized to develop a formulation for sectionalizing
the network [Fie75]. The distribution network is represented as a graph G = (V,E) as the buses and lines
15
are presented by vertices 𝐕 ∈ ℕ𝑁𝐵 and edges 𝐄 ∈ ℕ𝑁𝐿 respectively. Graph G has a corresponding
Laplacian matrix 𝐋 ∈ ℤ𝑁𝐵 × ℤ𝑁𝐵in which the diagonal arrays are the total number of edges connecting to
the vertices and the off-diagonal arrays represent the number of edges connecting vertices multiplied by ”
1 ”. The Laplacian matrix of a graph can be decomposed into its eigenvalues 𝜆1 ≤ 𝜆2 ≤ ⋯ ≤ 𝜆𝑁𝐵 as
shown by (1). Here, Q is the eigenvector matrix and Λ is the diagonal matrix of eigenvalues.
1)2.( T
L = QΛQ
A sample graph is represented in Figure 2-1, and the spectra of this graph which represent its
connectivity are given in Figure 2-2. For a connected undirected graph, the smallest eigenvalue of the
Laplacian matrix is always zero and the second smallest eigenvalue, also called Fiedler eigenvalue,
represents the algebraic connectivity of the graph [Fie73]. The number of zero eigenvalues is equal to the
number of isolated subgraphs in a graph [Fie75]. In the graph shown in Figure 2-1b, the first two
eigenvalues are zero as shown in Figure 2-2. One extreme case is the fully connected graph where the first
eigenvalue will be zero and other eigenvalues are equal to the number of vertices in the graph. The other
extreme case is the fully disconnected graph where all the eigenvalues are zero, so the number of sub-
graphs is equal to the number of vertices in the graph.
Figure 2-1 The sample connected graph (a). The sample sectioned graph (b).
4
5
6
1 2
3
4
5
6
1 2
3
(a) (b)
16
Figure 2-2 The spectra of the sample graph.
2.2 Problem Formulation
In this section, a new formulation is proposed to sectionalize the distribution network into
microgrids and the Fielder eigenvalue is used as a measure to determine the number of sub-networks in a
network. The problem is formulated as a bi-level mixed integer nonlinear problem (MINLP) [Bar98] in
which, the binary variables were associated with each edge of the distribution network graph i.e.
distribution lines. If the binary variable is 1 the distribution line remains connected, otherwise, the line will
be disconnected to form microgrid. The objective of the upper-level problem is to procure the maximum
number of microgrids while the lower-level problem ensures the reliability of the formed microgrids.
Therefore, in the upper-level problem, the distribution lines are disconnected to increase the number of
islanded microgrids in the case of a disturbance in distribution network; however, the objective of the
lower-level problem is to expand the distribution network within the formed microgrids to ensure that the
required level of reliability at demand buses is satisfied considering limited budget for such expansions.
Employing the duality theory, the presented bi-level problem is transformed into a single-level problem in
which the lower-level problem is considered as a constraint for the upper-level problem [Arr10]. The
expansion decisions are further validated to avoid over-investing on the distribution assets by minimizing
the cost of expansion plans considering the reliability constraints in the formed microgrids.
2.2.1 Network topology problem (Upper-level problem)
The upper-level problem minimizes the eigenvalues to maximize the number of microgrids in the
distribution network. The eigenvalue minimization problem has a convex feasible set and can be
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
1 2 3 4 5 6
Connected
Sectionned
Number of EigenvaluesE
igen
val
ue
17
represented as a semi-definite programming problem [Boy93]. In the upper-level problem (2.2)-(2.8), the
decision variables are the connection states of the distribution lines, which determine the PCCs in the
distribution network. In order to determine the decision variables, the eigenvalues of the Laplacian matrix
of the distribution network graph are minimized in an ascending order to ensure that 𝜆1 ≤ 𝜆2 ≤ ⋯ ≤ 𝜆𝑁𝐺 .
The sum of the weighted smallest eigenvalues – which are equal in number to the number of DERs – is
minimized in (2.2). In an undirected graph, the sum of eigenvalues of the Laplacian matrix of the graph is
equal to the sum of vertex degrees. In order to procure the maximum number of microgrids, the eigenvalues
of the Laplacian matrix of the distribution network graph should be zero in an ascending order. The first
eigenvalue is always zero. If the first two eigenvalues become zero, the distribution network will be divided
into two microgrids. Similarly, if the first three eigenvalues become zero, the distribution network will be
divided to three microgrids. Thus, the objective function of the upper-level problem is formulated to
procure zero eigenvalues in an ascending order, and the number of formed microgrids cannot exceed the
number of DERs. Here, the priority is to minimize the eigenvalues with lower orders. Minimizing the sum
of the eigenvalues does not guarantee that the eigenvalues will be zero in an ascending order. Therefore,
eigenvalues are weighted in the objective function of the upper-level problem.
Each eigenvalue is determined by multiplying the transposed matrix of eigenvectors, the Laplacian
matrix, and the eigenvector matrix as shown in (2.3). Thus, minimizing the eigenvalues will change the
Laplacian matrix, and consequently the topology of the distribution network. As indicated by (2.4), the
eigenvectors corresponding to each eigenvalue of the Laplacian matrix are orthogonal. The eigenvector
multiplication is a nonlinear but convex function. The eigenvector corresponding to each eigenvalue of the
Laplacian matrix are normalized to unit length as shown in (2.5). In (2.6) and (2.7), the relationship
between the Laplacian matrix of the distribution network and the decision variables associated with the
connectivity of distribution lines (edges of the associated graph) are given. Here, 𝑑𝑖𝑎𝑔(𝐿) represents the
diagonal elements of the Laplacian matrix 𝐿. In (2.8), the investment for adding the new distribution lines
is limited by the total investment budget (𝐵), and 𝑐𝑏,𝑏′ is the cost associated with adding a new
distribution line. Therefore, for the existing distribution lines, this parameter is set to zero. As the upper-
level problem minimizes the weighted eigenvalue spectra of the Laplacian matrix of the distribution
18
network graph – which also represents the connectivity of the distribution network graph – it will limit the
expansion of the distribution network as more distribution lines will increase the connectivity of the
network graph. The expansion of the distribution network is further limited by the available budget for
installing the distribution lines. It worth noting that other economic criteria such as ownership as well as the
PCC and controller costs were not incorporated in the presented formulation. Such criteria could be
addressed in the upper-level problem by adding constraints for the connectivity of certain vertices in the
distribution network graph. Moreover, the costs associated with PCC and controllers could be incorporated
in (2.8).
min ( 1)NG
bb
NB b (2.2)
1,....Tb b b b NB q Lq (2.3)
0 , 1,....Tb b b b b b NB q q (2.4)
1 , 1,....Tb b b b b b NB q q (2.5)
,( ) , 1,....b bb
diag u b b b b NB L (2.6)
, , 1,....bb b b bbl u l b b b b NB L (2.7)
, ,. , 1,....b b b bb
c u B b b b b NB (2.8)
2.2.2 Reliability evaluation problem (Lower-level problem)
The lower-level problem minimizes the expected operation cost of the autonomous microgrids while
maintaining the reliability requirements of the individual demands. The decision variables in the lower-
level problem (2.9)-(2.24) are the real and reactive power dispatch of the DERs and the demand served at
buses in the distribution network considering the determined topology in the upper-level problem. Here, the
decisions made on the connectivity of the distribution lines in the upper-level problem are constrained by
the reliability requirements at demand buses captured in the lower-level problem. This problem is presented
19
as linear programming (LP) problem. The objective function is the expected operation cost of the
autonomous microgrids which includes the operation cost of generating electricity and purchasing
electricity from the main grid, as well as the penalty associated with demand curtailment as shown in (2.9).
The real and reactive power injection at each bus of the distribution network is shown by (2.10) and (2.11),
respectively. At each bus, the available generation is the summation of the power generated by DER and
the imported power from the utility, if the bus is connected to the feeder. The real and reactive power
limitations of DERs are shown in (2.12) and (2.13) respectively. The expected energy not supplied (EENS)
for real and reactive demand at each bus is limited by (2.14) and (2.15) respectively. The EENS for the
demand at each bus is procured by multiplying the loss of energy probability (LOEP) and the total expected
energy consumed. As shown in (2.16), the admittance of distribution line incorporates the binary variable
representing its connectivity, which is determined by the upper-level problem, as well as the binary
parameter representing the line outage in different scenarios. The real and reactive power injection at each
bus is linearized in (2.17) and (2.18). The constraints (2.19), (2.20) and (2.21) present the linearized
formulation for real, reactive and apparent power transmitted through the distribution line [Kho12]. The
power flow in the distribution line is limited by the line capacity as given in (2.22). The voltage magnitude
and phase angle at each bus in the distribution network are limited by (2.23) and (2.24), respectively.
, ,
, , , , ,, , ,
,
( )mint s t s
i g
NS NTs t s t s t s D s d s
t c i g g bi b t b tP P s t i g
F P c P VOLL P P
(2.9)
s.t.
b b
inj ,st ,s t ,s d ,sg i b,t b,t
g GR i GG
t ,sg .P P P PUZ
(2.10)
b b
inj ,st ,s t ,s d ,sg i b,t b,t
g GR i GG
t ,sg .Q Q Q QUZ
(2.11)
, , ,min max .t s t s t s
i ii i iP UX P P UX (2.12)
, , ,min maxt s t s t si ii i iQ UX Q Q UX (2.13)
, ,, ,
0NT NS
s D s d st bb t b t
t s
P P EENS (2.14)
20
, ,, ,
0 d s D sb t b t
Q Q (2.15)
2 2 2 2
t ,s t ,sb,b b,b b,b b,bb,b b,bt ,s t ,s t ,s
b,b b,b b,bb,b b,b b,b b,b
r .UY .u x .UY .uy g Jb J
r x r x
(2.16)
12 1inj ,s
b,t
NBt,s t ,s t ,s t ,s t ,s t ,s t ,s t ,s
b b,b b bb,b b b,b bb b b
P BV G G V V
(2.17)
12 1inj ,sb.t
NBt,s t ,s t ,s t ,s t ,s t ,s t ,s t ,s
b b,b b bb,b b b,b bb b b
GQ V B B V V
(2.18)
, , , , , . ,, , ,
( ) ( )t s t s t s t s t s t s t sbbb b b b b b b b
GPL V V B (2.19)
, , , , , , ,, , ,
( ) ( )t s t s t s t s t s t s t sb bb b b b b b b b
QL B V V G (2.20)
, , ,,, , ,
t s t s t sb bb b b b b b
SL PL QL (2.21)
, max,,
t sb bb b
SL SL (2.22)
min , maxt sb
V V V (2.23)
min , maxt sb
(2.24)
2.2.3 Microgrid topology validation problem
Minimizing the sum of the eigenvalues that are equal to in number to the number of DERs will not
minimize the connectivity inside the formed microgrids. As the only constraint for limiting the number of
installed distribution lines is the limitation on the total budget, unnecessary distribution lines may be
installed. Installing the distribution lines will improve the reliability in the microgrids, however, such
decisions may not be economical once the reliability levels requested by demand entities are satisfied. To
address this issue, a microgrid topology validation problem is proposed to check if the added distribution
lines are necessary to satisfy the reliability requirements. Eliminating the unnecessary distribution lines
within the formed microgrids will not change the number of zero eigenvalues determined in the bi-level
optimization problem and consequently will not change the boundary of the formed microgrids. By
removing the unnecessary distribution lines within the microgrids, the sum of the non-zero eigenvalues is
21
decreased. Therefore, the algebraic connectivity is decreased within microgrids while the reliability
requirements are satisfied. The objective of the microgrid topology validation problem is given in (2.25), in
which the investment cost of adding new distribution lines and the operation cost of the system are
minimized. As this problem checks for the unnecessary installation of the distribution lines, the decision
variables in this problem are limited by the procured decisions (𝑢′𝑏,𝑏′) of the bi-level optimization problem
(2.2)-(2.24) as shown in (2.26). This means that this problem seeks to validate the decisions made in the bi-
level problem. The objective function for this problem is subjected to set of constraints (2.10)-(2.24).
, ,
,
, , , , ,, , ,, ,
, , '
( ) .mint s t s
i g b b
NS NTs t s t s t s D s d s
t c i g g b b b b bi b t b tP P u s t i g b b b
F P c P VOLL P P c u
(2.25)
b,b b,bu u b,b , b b (2.26)
2.2.4 Solution Methodology
The presented bi-level MINLP problem (2.2)-(2.24) is expressed in the general form by (2.27)-(2.32). By
employing the duality theory, the presented problem is transformed into a single-level problem (2.33)-
(2.40). The other approach for such transformation is to use the Karush Kuhn Tucker (KKT) conditions for
the lower-level problem; however, using the duality theory will generate less number of the binary-to-
continuous variable multiplications, which reduces the computation complexity and solution time. The
decision variables in the upper-level problem include the connectivity Ι and the eigenvalue and
eigenvectors η . Here, (2.27) minimizes the weighted eigenvalues in an ascending order as shown in (2.2).
The binary variables are the decision vectors representing the connectivity of distribution network
including the decisions on the location of PCCs and the installation of new distribution lines and the
continuous variables are the eigenvalue and eigenvectors. Set of constraints (2.28) represent the equality
constraints in the upper-level problem including the equality constraint from which the eigenvalues of the
Laplacian matrix are determined (2.3), the properties of the eigenvectors (2.4) and (2.5), and the
relationship between the connectivity of the distribution network graph and its Laplacian matrix (2.6),
(2.7). Constraint (2.8), which represents the limited budget for installing new distribution lines, is presented
by (2.29). The lower-level problem is shown in (2.30)-(2.32) in which the objective function (2.9) captures
22
the binary variables and re-written as (2.30), where vector d represents the parameters in the objective of
the lower-level problem. The set of equality constraints (2.31) represent the equality constraints (2.10),
(2.11), (2.16)-(2.21). Similarly, the set of inequality constraints (2.32) represents the inequality constraints
(2.12)-(2.15), and (2.22)-(2.24). The variables in this problem include the decision variables correspond to
connectivity represented by Ι , the eigenvalues and eigenvectors represented by η , the continuous
variables in the lower-level problem represented by α , and the Lagrange multipliers of the equality and
inequality constraints in the lower-level problem represented by π and τ .
,
minΙ η
η (2.27)
Subjected to:
( , ) 0,1g Ι η Κ Ι (2.28)
( )f Ι (2.29)
min T
αd α (2.30)
Subjected to:
( ) ( ) A B 1Ι α b π (2.31)
( ) ( ) C D 2Ι α b τ (2.32)
By employing the duality theory, the presented problem in (2.27)-(2.32) is transformed into a single-level
problem as shown in (2.33)-(2.40), where the lower-level problem is considered as a set of constraints for
the upper-level problem [Fie73]. The objective is the same as the objective of the upper-level problem,
which minimizes the eigenvalues of the distribution network graph in an ascending order. Here, the
objective and constraints of the upper-level problem are shown in (2.33)-(2.35) and the lower-level
problem (2.30)-(2.32) is shown as a set of constraints (2.36)-(2.40). The constraints (2.36) and (2.37) are
the constraints of the primal lower-level problem, while (2.38) and (2.39) are the constraints correspond to
the dual representation of the lower-level problem. The strong duality condition, which is shown in (2.40),
23
holds once the lower-level problem is linear and therefore the duality gap is zero. Here, the objective of the
primal lower-level problem is equal to the objective of the dual lower-level problem. The binary to
continuous terms in (2.40) are further linearized by (2.41)-(2.44).
min ( )Ι,η,α,π,τ
η (2.33)
Subjected to:
( , ) 0,1g Ι η Κ Ι (2.34)
( )f Ι (2.35)
( )A B 1Ι α b (2.36)
( )C D 2Ι α b (2.37)
T TB D π τ d (2.38)
τ 0 (2.39)
[ ( )] [ ( )]T T TA C 1 2d α π b Ι τ b Ι (2.40)
The equivalent single-level problem has several binary-to-continuous nonlinear terms. Such terms appear
in the AC power flow formulation and in the dual form of the lower problem. In (2.17)-(2.20) the binary
decision variables for the connectivity of the distribution network and installation of the distribution lines
are integrated into the admittance matrix as shown in (2.16).
The nonlinear binary-to-continuous terms illustrated in (2.41) are linearized by presenting the constraints
(2.42)-(2.44). Two auxiliary nonnegative variables, Φ and Ψ are employed and M is an arbitrary large
number.
, 0,1 (2.41)
(2.42)
0 M (2.43)
24
0 (1 )M (2.44)
The other sources of nonlinearities are the constraints that determine the eigenvalues. These
constraints form convex feasibility sets and an optimal solution of the presented MINLP is procured using
convex optimization approaches. Several approaches may be employed to solve the presented convex
MINLP problem. One of these methods is generating and successively improving the outer approximations
in the neighborhood of a set of optimal solutions for the MINLP. By utilizing the linear outer-
approximation and introducing auxiliary variables, the MINLP is reformulated into MIP problem that could
be solved by branch and bound [Ley01] or cutting plane techniques [Wes02]. Therefore, the presented
problem could be solved utilizing solvers (e.g. Alpha-ECP, BARON, and SCIP) that employ similar
methodologies [Taw04],[Wes03],[Ach09],[Vig13],[Bus14]. Finally, the decisions for adding new
distribution lines will be validated in the microgrid topology validation problem. This problem is
formulated as a MIP problem in which the decision variables are the installation states of the distribution
lines. The objective of this problem is to minimize the installation cost of the distribution lines and the
expected operation cost of the microgrids subjected to generation and distribution network constraints and
the reliability requirements of the demands.
2.3 Case Study
2.3.1 A Sample Meshed Distribution System
In this section, an electricity distribution network, which is composed of 20 buses, is illustrated in Figure
2-3. The presented distribution network is normally connected to the utility feeder while it is divided into
multiple microgrids fed by 5 DERs in the case of any disruption or disturbance in the distribution network.
The distribution network has 23 existing electricity distribution lines, and 20 electricity demands. Nine
electricity distribution lines are considered as candidates in the distribution network. Table 2-1 presents the
characteristics of the DERs and Table 2-2 shows the associated loss of energy probability (LOEP)
[Pra85],[Nii00],[Bas10],[Shaf15] at each bus which determines the priority of demands served in the
distribution network and calculated by (2.45). Table 2-3 describes the characteristics of the electricity
distribution lines including the forced outage rates (FOR), buses that are connected by each line, length and
25
the maximum capacity of the lines. The FOR associated with the feeder is 10.9 f/yr and the price of
electricity at the feeder is 8 ¢/kWh. Here, it is assumed that the available budget is sufficient for 1,000-
meter installation of the distribution lines. The states of the components are calculated based on the
availability of the components using two state Markov Chain process and Monte Carlo simulation. Here,
3000 scenarios are generated to represent the uncertainties in the system; including the generator and
distribution line random outages, and the demand uncertainties. Truncated normal distribution function is
used to represent the errors in load forecast, where the mean value is the forecasted volume and the
standard deviation is percentages of the mean values [Bil96], [Gre03]. Here, the standard deviation is 5%
from the mean value. The load profile is represented by the load duration curve which is divided into four
periods 1 2 3, , , and 4 with 10%, 25%, 60%, and 5% of the total annual hours. The scenario reduction
technique [Dup03],[GAMS] are employed to reduce 3000 scenarios to 12 scenarios. The voltage magnitude
and phase angles of buses are restricted between 0.95-1.05 per unit and (-π) to (π) respectively. The
resistance and inductive reactance of the distribution network cables are r = 0.092W /1000m and
x = 0.121W /1000m respectively. The equivalent single level MINLP is solved by SCIP 3.2.0 that employs
branch and bound algorithm, wherein the relaxed LP is tightened by cutting plane and domain propagation.
In order to handle the computation burden, sets of distribution lines that satisfy primal dual representation
of the reliability evaluation problem are procured and the set with minimum Eigenvalues is selected using
SCIP solver.
,,
s D sb b t b t
t s
LOEP EENS P (2.45)
1b bVOLL LOEP (2.46)
Table 2-1 Power generation unit characteristics
Unit Pmax1
(kW)
Pmax2
(kW)
Fc1
(¢/kWh)
Fc2
(¢/kWh)
Pmax
(kW)
Qmax
(kvar)
Qmin
(kvar)
FOR
(f/yr)
G1 1,200 600 8 15 1,800 900 -900 3.65
G2 900 450 10 28 1,350 625 -625 14.6
G3 1,000 500 10 24 1,500 750 -750 7.3
G4 900 450 9 14 1,350 600 -600 10.9
G5 750 325 12 25 1,075 550 -550 7.3
26
Table 2-2 Required LOEP at each bus
Bus bLOEP Bus bLOEP Bus bLOEP Bus bLOEP
1 0.03 6 0.005 11 0.05 16 0.03
2 0.005 7 0.001 12 0.09 17 0.07
3 0.03 8 0.001 13 0.05 18 0.01
4 0.01 9 0.07 14 0.08 19 0.02
5 0.02 10 0.04 15 0.05 20 0.03
Table 2-4 shows the reliability indices at each bus achieved in this configuration with no
microgrids or DERs. As shown here, the LOEPs of the loads are not within the required range and the
LOEPs exceed the acceptable level at buses 2, 4, 6, 7, 8, 11, 14, and 18. Considering the value of lost load
as defined in (2.46) in $/kWh, the expected annual operation cost of the distribution network is $1.773M.
The total expected demand and the total expected demand curtailment are 17,361 MWh and 2,247 MWh
respectively. The following two cases are considered,
1) Case 1 – heterogeneous LOEP requirement for demands
2) Case 2 – non-heterogeneous LOEP requirement for demands
Table 2-3 Distribution line characteristics
ID FOR
(f/yr)
Length
(m)
(kVA) ID
FOR
(f/yr)
Length
(m)
(kVA)
1 3.65 220 800 17 7.3 90 800
2 7.3 140 1,200 18 3.65 440 1,700
3 3.65 90 800 19 3.65 280 1,200
4 7.3 210 880 20 7.3 260 1,100
5 3.65 120 1,000 21 3.65 200 1,200
6 3.65 80 800 22 3.65 300 1,200
7 3.65 140 1,700 23 7.3 210 900
8 7.3 270 1,200 24 3.65 190 900
9 7.3 350 1,200 25 3.65 460 2,000
10 3.65 200 1,200 26 3.65 260 800
11 3.65 300 1,200 27 7.3 500 1000
12 3.65 120 1,200 28 7.3 450 900
13 7.3 110 1,200 29 3.65 390 900
14 3.65 160 2,000 30 7.3 360 1,000
15 7.3 220 800 31 3.65 500 800
16 7.3 120 1,000 32 3.65 460 1,000
Table 2-4 The LOEP in distribution network with no DER and microgrid
Bus Bus Bus Bus
1 0.015 6 0.015 11 0.342 16 0.015
2 0.015 7 0.015 12 0.022 17 0.019
3 0.022 8 0.015 13 0.019 18 0.015
4 0.015 9 0.019 14 1 19 0.015
5 0.02 10 0.019 15 0.015 20 0.015
maxSL maxSL
bLOEP bLOEP bLOEP bLOEP
27
Figure 2-3 The proposed electricity distribution network
2.3.1..1 Case1: Heterogeneous LOEP requirement for demands
In this case, the values for the required LOEP for demands at each bus are given in Table 2-2. The
spectra of the graph associated with the distribution network are shown in Figure 2-5. The first five
eigenvalues are minimized and the remaining eigenvalues are determined by validating the microgrid
topology. In this case, as the 5 smallest eigenvalues of the corresponding Laplacian matrix are zero, placing
the breakers at PCCs would help to divide the distribution network into five autonomous microgrids as
shown in Figure 2-4. Here, microgrid M1 is composed of buses {1, 2, 3, 8, 9, 10, 13, 16} which is supplied
by the G1. Microgrids M2, M3, M4, and M5 are composed of buses {5, 7}, {6}, {4, 12, 14, 15, 17}, and
{11, 18, 19, 20} which are respectively supplied by G2, G3, G5, and G4. The PCCs for microgrids M1 and
M3 are located on lines L4 and L7. The PCC between microgrids M1 and M2 is located on lines L9 and the
PCC between microgrids M2 and M4 is located on line L10. Moreover, the PCCs between microgrids M1
and M4 are located on lines L8, L17, and L20. The PCCs between microgrids M5 and M1; and between
microgrids M5 and M4; are located on lines L15 and L16, respectively. To address the reliability
requirements, distribution lines L32 and L29 with a total length of 850 meters were installed in the existing
distribution network. Table 2-5 shows the reliability indices achieved in this configuration. As shown in
this table, the LOEPs of the loads are within the required range, while the LOEP reached but not exceeded
G3
G2
7
4
10 2
186
9
3
5
L1
L2
L3 L4
L5 L6
L7
L8 L9
L10
L11
G4
G517
14
20 1211
18
1619
13
15
L12
L13
L14
L15
L16
L17
L18
L19
L20
L21 L22
L23
Feeder
L24
L25
L26
L27L28
L29
L30
L31
L32
Existing Lines
Candidate Lines
G1
28
the maximum acceptable level at bus 4. Considering the value of lost load as defined in (2.46) in $/kWh,
the expected annual operation cost of the distribution network with microgrids is $1.682M which is 2.93%
less than the operation cost of the distribution network in normal condition (2.$1.733M). Although more
expensive DER units serve demands within each microgrid, the reduction in the demand curtailment
reduces the operation cost. Hence, the formation of microgrids with DERs improves the operation cost of
the distribution network while the PCCs will form microgrids in the event of disruptions or failures in the
distribution network.
Table 2-5 The LOEP of demand at each bus in case 1
Bus bLOEP Bus bLOEP Bus bLOEP Bus bLOEP
1 0.0006 6 0.002 11 0.003 16 0.0006
2 0.0006 7 0.00005 12 0.09 17 0.016
3 0.0006 8 0.0006 13 0.0006 18 0.004
4 0.01 9 0.0006 14 0.008 19 0.003
5 0.00005 10 0.0006 15 0.008 20 0.003
The presented framework provides the necessary tool to define boundaries of each microgrid. The
procured microgrids are able to satisfy the reliability requirements of the demands in island mode. The total
expected demand curtailment is 112 MWh while the total expected annual demand is 17,361 MWh.
Forming the microgrids would assign the DERs to serve certain demands by islanding which leads to
heterogeneous LOEPs in the distribution network. Hence the presented approach ensures the reliability
requirements of the demands within the distribution network, while maximizing the number of microgrids
in the distribution network.
29
Figure 2-4 Electricity distribution network Case 1
Figure 2-5 The spectra of the graph associated with distribution network
2.3.1..2 Case 2: Non-heterogeneous LOEP requirement for demands
In this case, the new topology of the distribution network is procured considering a uniform
required LOEP for the demands. The required LOEP is the lowest required LOEP shown in Table 2-2. The
spectra of the distribution network graph are shown in Figure 2-5. Since the 3 smallest eigenvalues of the
related Laplacian matrix are zero, the distribution network is sectionalized into 3 autonomous microgrids as
shown in Figure 2-6. Here, microgrid M1 is composed of buses {1, 2, 3, 4, 6, 8, 9, 10, 13, 15, 16, 17}
which are supplied by the G1 and G3. Microgrids M2 and M3 are composed of {5, 7}, {11, 12, 14, 18, 19,
20} which are respectively supplied by G2; G4 and G5. The PCC between microgrids M1 and M2 are
placed on lines L9 and L10. In addition, PCCs between microgrids M1 and M3 are located on lines L15,
L17, and L21. To address the reliability requirements, the distribution lines L28 and L29 with a total length
G3
G2
7
10 2
18 9
5
L1
L2
L3
L5 L6
L11
G4
G517
14
11
18
16
15
L12
L13
L14L18
L19
L21 L22
L29
L32
19
L4
L7
L16
L17
L20
6
Line
PCC
L9
L10
13
12
20
L15
Boundary
3
G1M1
M2
M3
M4
M5
L23
4
L8
0
1
2
3
4
5
6
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Distribution Network
Case 1
Case 2
Number of Eigenvalues
Eig
env
alu
e
Eigenvalues corresponding to microgrids network topologies
30
of 860 meters were installed in the existing distribution network. The required reliability for all demand
entities is ensured as shown in Table 2-6. Here the required LOEPs for buses 5 and 7 are the same as those
in Case 1 because the topology of microgrid M2 in this case, is the same as microgrid M2 in Case 1. With
the value of loss load defined in (2.46) in $/kWh, the expected cost of the distribution network is $1.492M
in this case which is 11.24% less than that in Case1. Compared to Case 1, some of the demands require
lower LOEP in Case 2 which leads to larger microgrids (less zero eigenvalues for the graph spectra that
indicates more interconnected distribution network) as shown in Figure 2-6. The total expected demand
curtailment is 709.56 kWh. The LOEP for each demand is calculated in Table 2-6.
Table 2-6 The LOEP of demand at each bus in Case 2
Bus bLOEP Bus bLOEP Bus bLOEP Bus bLOEP
1 0 6 0 11 0 16 0
2 0.0001 7 0.00005 12 0.0006 17 0.00009
3 0 8 0 13 0 18 0
4 0 9 0 14 0 19 0
5 0.00005 10 0 15 0.00009 20 0
Figure 2-6 Electricity distribution network in Case 2
2.3.2 The Modified IEEE 123-bus system
The IEEE 123-bus system [IEEE123] is fed by the utility feeder in normal condition as shown in
Figure 2-7. The network will sectionalize into several microgrids served by 10 DERs in the case of
L17
L15
G3
G2
7
10 2
186
9L1
L2
L3
L5 L6
L7
L11
G4
G5
1714
2012
11
1619
13
15
L12
L13
L14
L16
L18
L19
L20
L22
L23
L28
L29
L9
3
5
Line
PCC
Boundary
L10
L21
18
G1
M1
M2
M3
L4
L8
4
31
disturbances in the network. Here, 8 distribution lines are considered as candidates in the distribution
network to satisfy the required LOEP at the demand buses. The required LOEPs at demand buses are given
in Table 2-7. As the 9 smallest eigenvalues of the corresponding Laplacian matrix of the distribution
network graph are zero, the distribution network is sectionalized into 9 autonomous microgrids. Microgrid
M1 is composed of buses {1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 34, 149} which are supplied
by the DER on bus 7 with the FOR of 10.9 f/yr. Microgrid M2 is composed of buses {18, 19, 20, 21, 22,
23, 24, 35, 36, 37, 38, 39, 135} which are supplied by the DER on bus 21 with the FOR of 3.65 f/yr. The tie
switch between buses 18 and 135 is closed and the PCC between microgrids M1 and M2 is placed on line
L13, which connects buses 13 and 18. Microgrid M3 is composed of buses {25, 26, 27, 28, 29, 30, 31, 32,
33, 250} which are supplied by the DER on bus 26 with the FOR of 7.3 f/yr. Here, the PCC between
microgrids M2 and M3 is placed on line L24 that connects buses 23 and 25. Microgrid M4 is composed of
buses {40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 151} which are supplied by the DER on bus 44 with
the FOR of 10.9 f/yr. Here, the tie switch between buses 151 and 300 is open and the PCC between
microgrids M2 and M4 is placed on line L36 that connects the buses 35 and 40. Microgrid M5 is the largest
designated microgrid that is composed of buses {52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 67, 72, 73, 74, 75,
76, 77, 78, 79, 80, 81, 82, 83, 84, 85, 86, 152, 160, 610} which are supplied by the DERs on buses 53 and
76 with the FOR of 10.9 f/yr and 14.55 f/yr respectively. Here, the tie switch between buses 60 and 160 is
closed and the PCC between microgrids M1 and M5 is located at the tie switch that connects buses 152 and
13. Microgrid M6 is the smallest designated microgrid that is composed of buses {62, 63, 64, 65, 66}
which are supplied by the DER on bus 62 with the FOR of 3.65 f/yr. Here, the PCC between microgrids
M6 and M5 is on line L61 that connects buses 60 and 62. Microgrid M7 is composed of buses {87, 88, 89,
90, 91, 92, 93, 94, 95, 96} which are supplied by the DER on bus 89 with FOR of 7.3 f/yr. Here, the PCCs
between microgrids M5 and M6 are on line L86 that connects buses 86 and 87, and on the tie switch that
connects buses 54 and 94. Microgrid M8 is composed of buses {68, 69, 70, 71, 97, 98, 99, 100, 450} which
are supplied by the DER on bus 98 with the FOR of 3.65 f/yr. To satisfy the reliability requirements at the
load point on bus 69, a 200-meter distribution line connects buses 70 and 100. The PCCs between
microgrids M5 and M8 are located on lines L66 and L68 that connect bus 67 to buses 68 and 97
respectively.
32
The FOR of the line L69 that connects buses 68 and 69 is 7.3 f/yr, which leads to multiple outages in the
operation horizon. Such high FOR results in unacceptable LOEP of the demand on bus 69 (0.06) before
forming microgrid M8. The required LOEP for the demand on buses 68 and 69 are 0.06236 and 0.0078
respectively. The network expansion in microgrid M8 by installing the distribution line that connects buses
70 and 100, will further reduce the LOEP of demand on bus 69 from 0.06 to zero, and the LOEP of demand
on bus 68 from 0.349 to 0.059. Thus, network expansion within autonomous microgrids further improves
the reliability of supplied energy.
Finally, microgrid M9 is composed of buses {101, 102, 103, 104, 105, 106, 107, 108, 109, 110, 111, 112,
113, 114, 197, 300} which are supplied by the DER on bus 108 with the FOR of 7.3 f/yr. Here, the PCC
between microgrids M8 and M9 is located on the tie switch that connects buses 197 and 97, and the PCC
between microgrids M4 and M9 is located on the tie switch that connects buses 151 and 300.
By considering 3.5% annual increase in demand in the 10-year operation horizon, and the value of
lost load ($/kWh) shown in (2.46), the expected operation cost of the distribution network with microgrids
is $45.6M. To illustrate the merit of forming microgrids in contingencies, the presented approach is
compared with the network reconfiguration technique, which is discussed in [Bor12]. Here, the existing tie-
switches within the IEEE 123 bus test system are utilized to reconfigure the distribution network
considering the outages in the distribution lines. The operation cost of the system with network
reconfiguration capability is $50.9M, which is 11.7% more than the operation cost of the distribution
network with microgrids. Similar to the first case study, forming microgrids supplied by DER units,
enhances the reliability and reduces the operation cost of the distribution network.
33
Table 2-7 Required LOEP of demands in IEEE-123 bus system
Bus bLOEP Bus bLOEP Bus bLOEP Bus bLOEP
1 0.03332 32 0.00516 63 0.01042 94 0.00104
2 0.11646 33 0.00984 64 0.0037 95 0.00516
3 0.2076 34 0.07228 65 0.00262 96 0.12146
4 0.11782 35 0.00056 66 0.01136 97 0.00084
5 0.0329 36 0.00974 67 0.00012 98 0.02432
6 0.24462 37 0.00042 68 0.06236 99 0.0652
7 0.06376 38 0.0002 69 0.0078 100 0.02492
8 0.02628 39 0.00868 70 0.000742 101 0.0023
9 0.00134 40 0.02484 71 0.00624 102 0.0012
10 0.01234 41 0.00364 72 0.00354 103 0.021
11 0.05062 42 0.00952 73 0.00274 104 0.00004
12 0.07316 43 0.00004 74 0.00222 105 0.04598
13 0.01622 44 0.00176 75 0.0611 106 0.00002
14 0.0724 45 0.00242 76 0.00034 107 0.08878
15 0.04926 46 0.08068 77 0.04038 108 0.03392
16 0.00112 47 0.01546 78 0.0006 109 0.00076
17 0.00056 48 0.00892 79 0.00096 110 0.00598
18 0.06462 49 0.06266 80 0.00104 111 0.09222
19 0.00006 50 0.05122 81 0.00006 112 0.00512
20 0.00014 51 0.05926 82 0.00302 113 0.000006
21 0.0024 52 0.00666 83 0.02186 114 0.1334
22 0.00012 53 0.02476 84 0.04486 197 0.02056
23 0.00056 54 0.0151 85 0.05452 160 0.00116
24 0.00182 55 0.02636 86 0.10082 152 0.01408
25 0.00008 56 0.0066 87 0.00338 149 0.00106
26 0.00034 57 0.00056 88 0.01694 135 0.00107
27 0.05124 58 0.003 89 0.00014 300 0.00003
28 0.00034 59 0.0147 90 0.000364 450 0.00001
29 0.00038 60 0.00134 91 0.01478 350 0.00055
30 0.03862 61 0.0372 92 0.11858 250 0.00016
31 0.00454 62 0.00946 93 0.0014
34
Figure 2-7 Modified IEEE 123-bus test system
2.4 Summary
This chapter proposes a framework to sectionalize the distribution network into autonomous
microgrids, considering heterogeneous reliability for the demands. The expansion of distribution network
and formation of microgrids satisfy the reliability requirements in case of disruptions in the distribution
network. The presented problem is formulated as a bi-level optimization problem. In the upper-level
problem, the number of microgrids is maximized by minimizing the weighted sum of the eigenvalues of the
Laplacian matrix of the distribution network graph. In the lower-level problem, the expected operation cost
is minimized considering the reliability constraints for the demands at each bus. The procured microgrid
topologies are further validated by minimizing the expansion and operation cost within the microgrids. The
presented approach is evaluated by two case studies.
32
31
27
33
26
28
25
2930 250
24
22
23
21
20
1918
135
48
44
47
45
46
42
40
49
43
5051
151
35
37
3938
36
11
2
14
910
1149
8
12
713
34
4
3 5 6
16
15 17
95
96
93
94
91
92
89
90
87
88
86
82
83
84
81
80
85
79
7877
76
7574
7372
61061
6716060
57
5958
62
63
64
65
66
68
71
70
69
450
98
97
197
110
102
104
103
107
106
105
114
350
300
108
109
111 113110 112
56
152
555453
52
99
10041
35
Chapter 3
3RESILIENT OPERATION OF MULTIPLE ENERGY CARRIER MICROGRIDS
In this chapter, the energy resource interdependence in microgrids is represented by implementing energy
hubs, and preventive reinforcement strategies are offered to the microgrid operators following the
resiliency analysis. The contributions of this chapter are as follows:
• Modeling the disruption as a consequence of deliberate actions in multiple energy carrier
microgrids considering the interdependence of natural gas and electricity infrastructures
• Identify the critical and vulnerable components in multiple energy carrier microgrids
• Quantify the consequences of disruption by calculating the operation cost of multiple energy
carrier microgrid in contingencies
• Apply preventive reinforcement strategies to ensure the resilient operation of microgrids and
determine the reinforcement outcomes
The results can be used by microgrid planners and operators to identify critical components whose
reinforcement will improve the reliability and security measures within the microgrids. The chapter is
organized as follows: Section 3-1 introduces the basic components of the energy hub. Section 3-2 describes
the preventive reinforcement in cyber infrastructure to improve the resilience in the energy supply. Section
3-3 describes the problem formulation. Section 3-4 presents a case study, to show the effectiveness of the
proposed preventive reinforcement procedure, and section 3-5 presents the conclusion.
36
3.1 Energy Hub
Energy infrastructure provides various services to industrial, commercial, and residential consumers. The
optimal expansion planning and operation of various energy carriers are determined as a set of independent
problems without taking the interdependencies and interactions of the energy carrier networks into account.
Here, the nodes in the distribution network with energy conversion facilities are referred to as “energy
hubs”. The energy hub incorporates the interdependencies between natural gas, electricity and heat and
provides operational flexibility within active distribution networks as shown in Figure 3-1.
Here, the heat energy is produced as a bi-product of electricity generation using natural gas in
combined heat and power (CHP) units and curtailing the electricity demand at each hub would result in
curtailment of heat demand, as the electricity drives the pumps to facilitate heat transfer in the energy hubs.
Figure 3-1 Energy hub with electricity, natural gas and heat
The interdependence of electricity and heat demands with electricity and natural gas resources
provided at each energy hub is shown in (3.1), where the electricity and natural gas comprise input vector
E , and the electricity and heat demand comprise output vector L . Here, matrix C is the forward coupling
matrix which represents the input-output energy conversion. The forward coupling matrix is composed of
coupling factors and is determined by characteristics and topology of the energy hub.
e e gas e ee
gash e h gas h
C C EL
EL C C
L EC
(3.1)
37
3.2 Preventive Reinforcement in Cyber Infrastructure
Multiple energy carrier microgrids ensure the continuity of service to critical demands in energy
infrastructure. Hence the resilience of multiple energy carrier microgrids against cyber threats leading to
disruptions in physical components in energy infrastructure is addressed. Multiple energy carrier
microgrids cannot tolerate excessive restoration times, as the recovery and restoration is not considered as a
viable option for the mission critical facilities served by microgrids. Hence the resilience is defined as the
ability of the system to survive attacks without suffering from any service interruptions and the objective is
to provide a solution framework to ensure the continuity of service in multiple energy carrier microgrids.
The set of preventive actions that is required for resilient operation of microgrid, is defined through the
proposed reinforcement procedure.
More than half of the cyber threads are in energy sector [ICS13]. The most frequent cyber threats
are authentication, denial of service, and buffer overflow [ICS13]. Here, it is assumed that the objective of
the attacker is to increase the service interruption by disrupting the physical components through cyber-
attacks. The physical components of the microgrid receive encrypted data packets which contain the control
signals from the microgrid controller. Hence decryption and manipulating the encrypted data packets would
result in the disruption of the physical components. Figure 3-2 illustrates disruption of electricity
distribution line by decrypting the control signals sent from the microgrid controller. Here, the disruption
occurs if the encrypted data is decrypted and manipulated by the attacker. Moreover, it is assumed that the
data packets correspond to each component is encrypted with an existing encryption technique e.g. secured
hash functions (SHA-x) [Mir05], or advanced encryption standard (AES) [Dae02]. However, other
encryption techniques [Den82] could also be applied in the proposed formulation. The higher encryption
strength is achieved by the increase in the encryption cost which eventually leads to higher disruption cost
of the physical component. The cost of providing higher encryption strength is dependent on the hardware
implementations such as processing modules and communication channels. In this chapter, an encryption
cost is assigned to specific encryption strength provided by practical encryption techniques. For instance,
the encryption cost of using AES-256 to encrypt a data packet is more compared to the encryption cost of
using AES-192 considering the same performance measures. While the encryption strategy to improve the
38
encryption strength of the data packets associated with the physical components are discussed in this
chapter, the incorporated encryption technique is out of the scope of this chapter. The presented model
considers an initial encryption strength for the data packets corresponding to each physical component. The
new encryption strategy is to improve the encryption strength for the certain components which are
identified by the solution of the bi-level optimization problem employed. Hence, an improved encryption
strategy decreases the vulnerability of microgrid against disruptions and increase the cost of decryption,
data manipulation, and disruption of the physical components. In other words, each stage of the preventive
reinforcement corresponds to the encryption strategy chosen in the microgrid. The encryption strategy
taken, includes the improvements in the encryption strength of the selected physical components of the
microgrid.
Figure 3-2 Disruption of service in multiple energy carrier microgrids
Here, it is assumed that different types of physical components in the microgrid require different numbers
of data packets. The number of the data packets is dependent on the number of controllable variables in
each component. For instance, it is assumed that generators require larger number of data packets for
monitoring and control compared to the distribution lines.
The preventive reinforcement presented here is defined in several stages, where the microgrid resilience
at each stage is dependent on the encryption strategy incorporated for monitoring and control of the
microgrid. The solution to the bi-level optimization problem at each stage procures the new encryption
39
strategy and the iterative procedure continues until the desired level of resilience is established. The desired
level of resilience is determined by the microgrid planner or operator. To quantify the resilience of the
microgrid against disruptions, a resilience index (r) shown in (3.2) is defined as the exponential of negative
ratio of the increase in operation cost of multiple energy carrier microgrid as a result of disruptions to the
budget of the attacker causing such disruptions. Here, and 0 are operation costs of disrupted and non-
disrupted multiple energy carrier microgrid. Thus low resilience index indicates that the multiple energy
carrier microgrid is highly vulnerable against disruptions.
0( - )( )
e Mr
(3.2)
3.3 Problem Formulation
The operation of multiple energy carrier microgrids exposed to disruptions is presented as a bi-level
optimization problem. The upper level problem maximizes the operation cost of the system subjected to the
limitation of resources to trigger disruptions while the lower level problem minimizes the operation cost of
the multiple energy carrier microgrid under attack. Further, the MIBLP problem is re-formulated into a
single-level mixed integer programming (MIP) problem employing the duality theory.
The objective function of the upper level problem, expressed by (3), maximizes the operation cost while
the decision binary variables are the outage statuses of generation units, natural gas pipelines, and power
distribution lines. The operation cost of system includes the penalty cost for the load curtailments, hence
maximizing the operation cost of the system by the attacker would lead to maximizing the curtailed
demand and the generation cost in the microgrid. The resources for triggering outages in various
components of microgrid are limited as indicated in (3.4). In bi-level optimization problem presented here,
the lower level problem and its constraints are considered as constraints for the upper level problem
[Bar98]. The objective function of the lower level problem is the operation cost of the microgrid which
includes the operation cost of generating electricity and heat energy from natural gas, as well as the penalty
cost for electricity and heat demand curtailment as shown in (3.5). The real and reactive power injection is
shown by (3.6) and (3.7), respectively. For the sake of simplicity the parameter e eC which represents the
40
efficiency of the transformer is considered as unity. The real and reactive power limits for generation units
are formulated in (3.8) and (3.9) respectively. The curtailed real and reactive demand at each energy hub is
less than the total real and reactive demand as presented by (3.10) and (3.11) respectively. The admittance
of each power line shown in (3.12) incorporates the binary variable representing the outage status of the
component. The real/reactive power injection used for ac power flow is linearized in (3.13) and (3.14).
Equations (3.15), (3.16) and (3.17) present the linearized formulation for real, reactive and apparent power
transmitted through the distribution line. Here, ξ is an auxiliary parameter, which is dependent on the load
power factor as calculated in [Kho12]. The limitation on apparent power flow is imposed by distribution
line capacity as given in (3.18). The limitations on voltage magnitude and voltage phase angle in electrical
networks as well as the natural gas pressure within natural gas distribution network are presented in (3.19),
(3.20) and (3.21), respectively. The minimum and maximum capacity for natural gas resources are shown
in (3.22). The interdependence of electricity and heat demands is represented in (3.23), where K is a large
number, which indicates that the heat distribution facility requires electricity to transfer the heat. Thus,
electric demand curtailment at each energy hub would lead to heat demand curtailment. The heat demand is
served by the electricity bi-product or burning natural gas in heater as shown in (3.24). The natural gas
balance at each energy hub is presented by (3.25), in which the injected natural gas at each hub is equal to
the withdrawn volume by the interconnected pipelines as well as the energy hub to serve the electricity and
heat demand. The natural gas is converted into electricity and heat using the respective conversion factors.
Here, any outage in natural gas supply will result in the outage of the supplied generator. The constraint on
natural gas flow in distribution pipelines is given in (3.26). Here, the dependency of the natural gas flow on
temperature which is discussed in [Mar12] is ignored and the linear dependency of natural gas flow
between two interconnected energy hubs and their pressures are expressed in (3.27).The served heat
demand is lower than the total heat demand at each energy hub as shown in (3.28).
,
,, ,
, ,maxl
i j o p
li j o p
UX UY UZ
UX UY UZ (3.3)
Subject to:
41
NG NL NPl
i i l p pi l p
M UX M UY M UZ M (3.4)
, ,
, , , ,, , ,
, ,
,
( , , ) ( ) .( ) ( ) .( )mine h
i k
e d h dj j
l e e e D e d h h h D h di j o p c i i c k kj j j j
P P i j k j
P P
UX UY UZ F P VOLL P P F P VOLL P P
(3.5)
j j
inje e,di j j
i GG j D
P P P
(3.6)
j j
inje e,di j j
i GG j D
Q Q Q
(3.7)
min max(1 ) (1 )ei i i i iP UX P P UX (3.8)
min max(1 ) (1 )ei i i i iQ UX Q Q UX (3.9)
, , ,0 e D e d e Dj j jP P P (3.10)
, , ,0 e D e d e Dj j jQ Q Q (3.11)
2 2 2 2
1 1i ,o j ,o i ,o j ,oj ,o j ,o j ,o
j ,o j ,o j ,o j ,o
r .( UY ) x .( UY )y g Jb J
r x r x
(3.12)
12 1injj
NB
j j, j j ,o j o j,o j oo o j
P BV G G V V
(3.13)
12 1injj
NB
j j, j j ,o j o j,o j oo o j
GQ V B B V V
(3.14)
, , ,( ) ( )j o j o j o j o j oGPL V V B (3.15)
, , ,( ) ( )j o j o j o j o j oQL B V V G (3.16)
, , , ,j o j o j o j oSL PL QL (3.17)
max, ,j o j oSL SL (3.18)
min maxjV V V (3.19)
42
min maxj (3.20)
min maxj (3.21)
min maxs s sv v v (3.22)
,0 h d ejjP K P (3.23)
,0 ( ) / ( )
j j
h d i e i hgas h i gas e kj
i GG k HH
P C P C P
(3.24)
, ,( ) ( )
h eP Pk iA v f fj s p p k is GS p P p P k HH i GGC Cj f j t j j jgas h gas e
(3.25)
,max ,max, , ,.(1 ) .(1 )p p p
p pj o j o j of UZ f f UZ (3.26)
, ,
2 2
( )p j j o op pp pj o j o
j o
Cf K UZ f K UZ
(3.27)
, ,0 h d h Dj jP P (3.28)
The presented MIBLP problem formulation (3.3)-(3.28) is expressed in the general form by (3.29)-(3.34).
By employing duality theory, the presented problem is transformed to single-level MIP problem presented
by (3.35)-(3.42). The proposed bi-level problem can be transformed to a single level MIP problem using
KKT conditions or duality theory. Here, the duality theory is used to decrease the number of the binary-to-
continuous variable multiplications, computation complexity and solving time [Arr10]. The objective
function (3.3) is represented by (3.29) and the disruption resource allocation constraint (3.4) is presented as
(3.30), where * is the optimal solution of lower level problem (3.32). The binary variables in (3.31) are
the decision vectors representing the outage statuses of electricity generation and distribution components
as well as the natural gas pipelines. The lower minimization problem, with determined binary variables,
given in (3.32) represents (3.5) which minimizes the operation cost of the microgrid. Equations (3.33) and
(3.34) represent all the equality and inequality constraints shown in (3.6)-(3.28), respectively. Here and
43
are introduced as the dual variables associated with equality constraints (3.33) and inequality constraints
(3.34), respectively.
*max (U, )U
h (3.29)
Subject to: *( , ) 0g U (3.30)
0,1U (3.31)
min Td
(3.32)
Subject to: 1( )A U B b (3.33)
2( )C U D b (3.34)
In the presented single level MIP formulation (3.35)-(3.42), the objective is to maximize the operation
cost of the system by determining the disruption binary variables U , generation dispatch , equality
constraint dual variable , and inequality constraint variable . The constraints (3.36)-(3.37) are similar to
(3.30)-(3.31) and * is replaced by , which is the decision variable in the equivalent problem (3.35)-
(3.42). In the original formulation (3.29)-(3.34), the binary decision variables which are determined in the
upper level problem (3.29)-(3.31) are fixed for the lower level problem (3.32)-(3.34). However, in (3.38),
the dual form of the lower level problem is shown. In (3.39), it is stated that dual variables of the inequality
constraints of the lower level problem are nonnegative. The constraints (3.40) and (3.41) are the same as
(3.33) and (3.34), respectively. Here, U is the decision variable in the equivalent single level problem. The
equality constraint associated with strong duality is expressed in (3.42). Here (3.38) and (3.42) are
composed of several of binary-to-continuous variables multiplications, which are further linearized.
, , ,max (U, )
Uh
(3.35)
Subject to: ( , ) 0g U (3.36)
44
0,1U (3.37)
T TB D d (3.38)
0 (3.39)
1( )A U B b (3.40)
2( )C U D b (3.41)
1 2[ ( )] [ ( )]T T Td b A U b C U
(3.42)
To solve the current problem, all the nonlinear terms which are composed of binary-to-continuous
variable multiplication are transformed into linear forms. Such nonlinearities exist in three categories: a)
the ac power flow formulation, b) the natural gas flow formulation, and c) the dual form of lower problem
in the equivalent single level MIP problem. In the ac power flow formulation, the binary decision variables
for the distribution lines are embedded in the admittance matrix as shown in (3.12). Equations (3.13)-(3.16)
contain 10 binary-primal continuous variables multiplication terms, e.g. voltage as primal continuous
variable to binary variable associated with distribution line availability status in (3.13). The second
category is the natural gas flow formulations, as shown by equations (3.26) and (3.27). The last category is
the dual form of the lower level problem, which has binary-to-continuous variable multiplications in (3.42).
The linearized form of the term (3.43) is shown in (3.44)-(3.46) in which the lower and upper bounds of the
continuous variable are considered as zero, and a large number, K, respectively. Here, Φ and Ψ are
nonnegative continuous variables.
, 0,1U U (3.43)
(3.44)
0 K U (3.45)
0 (1 )K U (3.46)
45
The constraint (3.27) is the linear approximation of the original Weymouth equation (3.47) which
represents the relationship between the natural gas flow with the pressure at the inlet and the outlet of a
natural gas pipeline. Figure 3-3 shows the hyper plane represented by Weymouth equation, which is
decomposed into several smaller hyper planes. The linearization using Taylor series is valid only if the
difference in natural gas pressure between the inlet and outlet of the pipeline is assumed to be limited, i.e.
there is not significant pressure drop in the pipeline. This is a reasonable assumption for the short pipelines
used in microgrids. The limitation on the node pressure on the gas pipeline network shown in (3.21)
guarantees the accuracy of the approximation. The energy hubs within microgrids operate within specific
range of nodal natural gas pressure, e.g. between 55 and 56 bars. The Weymouth equation is linearized
around the initial point procured by solving optimal energy flow within the microgrid considering no
disruptions.
,
2 2.p
f j o p j oC (3.47)
Figure 3-3 Hyper plane for the Weymouth equation
3.4 Case Study
In this section, a multiple energy carrier microgrid composed of 10 energy hubs is illustrated in
Figure 3-4. The microgrid presented in the case study operates in island mode; however, the presented
preventive reinforcement procedure could be applied for microgrids operating in grid-connected mode by
incorporating the grid interconnection with respective price of electricity and disruption cost into the
formulation. The presented microgrid has 3 distributed generation CHP units with electrical and heat
46
efficiency of respectively 34% and 50% which consume natural gas to produce electricity and heat, 11
electricity distribution lines, 10 electricity demands, 6 heat demands and 5 natural gas pipelines. Table 3-1
presents the characteristics of the natural gas generators. Table 3-2 lists the electricity and heat demands
and their respective value of lost loads which determine the priority of demands. Table 3-3 describes the
characteristics of electricity distribution lines and the natural gas pipelines.
Figure 3-4 Multiple energy carrier microgrid.
Table 3-1 Natural gas power generation unit characteristics
Unit Pmax1
(kW)
Pmax2
(kW)
Fc1
(¢/kWh)
Fc2
(¢/kWh)
1
gas elecC
(SCM/kWh)
2
gas elecC
(SCM/kW
h)
Pmax
(kW)
G1 800 400 8 15 0.00789 0.0092 1,200
G2 1,200 600 10 28 0.00796 0.0102 1,800
G3 1,000 500 10 24 0.00726 0.0083 1,500
Table 3-2 Electricity and heat demand
Hub
ID
,e DP
(kW)
,e DQ
(kVar)
eVOLL
($/kWh)
,h DP
(MBtu)
hVOLL
($/MBtu) m
(bar)
1 80.7 40.4 10 95.23 1 55.34
2 113.0 56.5 8 111.11 1 55.10
3 161.5 80.8 10 142.85 1 55.11
4 242.3 121.2 10 126.98 1 55.00
5 290.7 145.4 10 158.72 1 55.01
6 323.0 161.5 10 158.72 1 55.37
7 80.7 40.4 100 - - -
8 161.5 80.8 10 - - -
9 323.0 161.5 10 - - -
10 323.0 161.5 20 - - -
47
Here, Cp for natural gas pipeline is a constant which is determined by temperature, diameter, gas
composition, length and friction [4]. The natural gas supply could deliver maximum volume of 50 Standard
Cubic Meter (SCM) at the source point. The voltage magnitude and phase angles of energy hubs within the
electrical distribution network are restricted between 0.95-1.05 per unit and -π to π respectively. The
resistance and inductive reactance of the distribution network cables are 0.028 /1000r ft and
0.037 /1000x ft respectively. The natural gas pressure at each hub is limited between 55 and 56 bars.
The equivalent single level MIP is solved by IBM ILOG CPLEX v12.6.0 software package.
Table 3-3 Distribution network characteristics (electricity and natural gas)
ID From
Hub
To
Hub
Length
(m)
maxSL
(KVA) Cp
maxf
(SCM)
L1 1 8 100 1,200 - -
L2 1 10 120 1,200 - -
L3 1 9 160 2,000 - -
L4 6 9 180 800 - -
L5 10 2 120 1,000 - -
L6 3 2 80 800 - -
L7 6 3 140 1,700 - -
L8 2 4 270 1,200 - -
L9 3 5 350 1,200 - -
L10 5 4 200 1,200 - -
L11 5 7 300 1,000 - -
P1 1 6 340 - 3 25
P2 2 3 80 - 2.82 25
P3 6 3 140 - 3 25
P4 3 5 350 - 2.82 25
P5 4 5 200 - 2.82 25
As stated earlier, an encryption cost is associated with a specific encryption strength provided by the
practical encryption techniques. For example, the encryption costs of a single data packet using AES-128,
AES-192, and AES-256 are assumed as $128, $256, and $512 respectively. Similarly, the encryption costs
correspond to the hash function with 128, 256, and 512 bits could be assumed as $128, $256, and $512
respectively. Any Further improvement in encryption strength achieved by cascading encryption using the
standard encryption techniques. The required resources to manipulate an encrypted data packet is assumed
to be 10 times of the encryption cost. Here, the encryption and disruption costs of the initial encryption
strength for each data packet are assumed as $128 and $1,280 respectively. The generator, natural gas
48
pipeline, and electricity distribution line use 7, 6, and 2 data packets. Hence the disruption costs are $8,960,
$7,680, and $2,560 for generator, natural gas pipeline and distribution line, respectively. Disruption
attempts with limited resources ($20,000) were considered in the following cases to evaluate the resilience
of the energy supply in the proposed microgrid:
• Case 1: No preventive reinforcement in the microgrid
• Case 2: First stage of preventive reinforcement, by increasing the disruption costs for the
vulnerable components determined in Case 1.
• Case 3: Second stage of preventive reinforcement by increasing the disruption costs for the
vulnerable components determined in Case 2.
• Case 4: Third stage of preventive reinforcement by increasing the disruption costs for the
vulnerable components determined in Case 3.
3.4.1 Case 1: No preventive reinforcement in the microgrid
In this section, the impact of disruptions on the energy infrastructure within a microgrid is
presented with no preventing reinforcement on the components. Applying the proposed methodology
highlights the vulnerable components of the microgrid against disruption as shown in Figure 3-5. The
attack in this case results in disruption of distribution lines L2, L3, L4, and L7 and the natural gas pipeline
P4. This will lead to the curtailment in electricity demands on energy hubs 2, 3, 4, 5, 7, 9, and 10 as well as
curtailment in heat demands on energy hubs 2, 3, 4, and 5. The operation cost increases from $195 in
normal operation to $26,275 in this case. The resilience index of multiple energy carrier microgrid in this
case is 0.2715. In this case, the electricity and heat demands on energy hubs 1 and 6 are served by
generation unit supplied by the natural gas. Since the electric demand on energy hub 9 is larger than that on
energy hub 8, the impact of disruption on energy hub 9 is higher; hence interruption on L3 and L4 will lead
to disruption of electricity supply on energy hub 9. On the natural gas distribution network, the natural gas
pipeline from energy hub 3 to energy hub 5 is the primary candidate for interruption and this interruption
will disrupt the operation of generator G2. Moreover, disconnecting electricity distribution lines L2 and L7
will result in forming four isolated electrical networks, i.e. {2, 3, 4, 5, 7, 10},{1,8}, {6}, and {9} within the
49
microgrid. The supply and demand balance is reached by curtailing the electricity and heat demand as
required. The total supplied electricity by G1 and G3 in this case is 559.4 kW which is equal to the served
demand in the islands composed of energy hubs {1, 8} and {6}, respectively. In this case, disruption of P4
left G2 without the gas supply, and the generation dispatch of G2 is zero; hence, the island formed by
energy hubs {2, 3, 4, 5, 7, 10} is left without any available electricity generation. As a result, the electricity
demand is curtailed in this island.
Figure 3-5 Microgrid operation condition in Case 1
3.4.2 Case 2: First Stage of Preventive Reinforcement
In this case, the encryption strategy is updated which indicates higher encryption strength for the data
packets utilized to control and monitor the vulnerable components identified by the solution of the bi-level
optimization problem in Case 1. The encryption cost for each vulnerable component is twice as that of the
initial encryption strength. This is represented by increasing the encryption cost of each data packet to $256
for the disrupted components. For instance, the encryption cost of each data packet corresponds to L2 is
increased from the initial cost of $128 to $256 so L2 is reinforced against disruptions in this case, by
spending $512 for the improved encryption strength. In this case, the interruption in distribution lines L1,
G1
Gas Supply
8 1
G3
6
G2
5
9
10 2 3
74ElectricityNatural Gas
Electricity OutageHeat OutageHeat
L1
L2
L3
L4
L5 L6
L7
L8
L9
L10 L11
P4
50
L2, L6, L9, L10 and L11 will lead to the interruption in electricity demand in energy hubs 2, 4, 7, 8, and 10
and the interruption in heat demand in energy hubs 2, 4 and 5. The operation cost will decrease by 20.2%
from $26,275 in Case 1, to $21,860 in this case and the resilience index is 0.3385 in this case.
Here, as the disruption cost of line L7 was doubled, this line is not a good candidate for disruption; instead,
interruption in L6 will result in the electric demand curtailment and consequently heat demand curtailment
on energy hub 2. Moreover, as energy hubs 7 and 8 are fed radially by lines L11 and L1 respectively,
interruption of these lines will lead to the electricity demand curtailment on energy hubs 7 and 8. As shown
in Figure 3-6, the disruption in L11 will lead to the load curtailment on energy hub 7 with very high value
of lost load (VOLL). Higher VOLL on energy hub 10 provokes the interruption of lines L2, L6 and L10
which feeds the demand on energy hubs 2, 4 and 10. As seen in the previous case, disruption in L9 isolates
energy hub 5 from energy hubs {1, 9, 6, 3}. In this case, several isolated energy hubs including {8}, {1, 9,
6, 3}, {10, 2, 4}, {5} and {7} are formed.
Figure 3-6 Microgrid operation condition in Case 2
G1
Gas Supply
8 1
G3
6
G2
5
9
10 2 3
74ElectricityNatural Gas
Electricity OutageHeat OutageHeat
L1 L2
L3
L4
L5
L6
L7
L8
L9
L10
L11
51
3.4.3 Case 3: Second Stage of Preventive Reinforcement
Similar to Case 2, the encryption strategy is updated based on the solution to the bi-level
optimization problem in Case 2. In this case, L1, L8, L10 and L11 were disrupted which led to the
electricity demand curtailment in energy hubs 4, 7, and 8 and the heat demand curtailment in energy hub 4.
The encryption strategy employed in the second stage of the preventive reinforcement reduced the
operation cost by 71.1% from that of Case 2. The operation cost in this case is $12,777 and the resilience
index is 0.5331. Here, the disruptions in electricity distribution lines L8 and L10 result in the electricity and
consequently the heat load curtailment in energy hub 4. In this case, the energy hub 4 is a good candidate
for the offender as L8 which feeds this energy hub has not been reinforced in the previous reinforcement
stages and still is vulnerable to attacks; hence, it requires lower resources for interdiction compared to other
reinforced distribution lines supplying energy hubs 2 and 3. As shown in Figure 3-7, the heat demand is
curtailed on energy hub 4 as a result of disruption in the electricity supply and the interdependence of heat
and electricity, as shown in (23). Similar to Case 2, interruptions in L1 and L11 will cause electricity
demand curtailment in energy hubs 8 and 7 respectively.
52
Figure 3-7 Microgrid operation condition in Case 3
3.4.4 Third Stage of Preventive Reinforcement
In this case, the electricity demand on energy hubs 1, 8, 9, and 10 were curtailed as a result of interruption
on L4, L5, and P1. The operation cost in this case is $12,770 which shows a slight decrease compared to
that of Case 3. The resilience index of multiple energy carrier microgrid is 0.5333 in this case. Moreover,
the energy hubs 1, 8, 9, and 10 in the microgrid were isolated as a result of the disruption in L4 and L5 and
G1 is shut down as a result of disruption in the pipeline P1. This will curtail the electricity and heat demand
on energy hub 1 because of disruption in both electricity and natural gas supplies. Although the curtailed
demand in this case is larger compared to that in Case 3, the operation cost is very close to that of Case 3
because of the higher VOLL for electricity demand on energy hub 7 which is not interrupted in this case.
3.4.5 Preventive reinforcement procedure
Increasing the number of iteration to reinforce the disrupted components will result in more costly
disruptions and less outages in the natural gas and electricity infrastructures. The preventive reinforcement
procedure will prioritize the components for further reinforcements to reduce the operation cost of the
G1
Gas Supply
8 1
G3
6
G2
5
9
10 2 3
74ElectricityNatural Gas
Electricity OutageHeat OutageHeat
L1L2
L3
L4
L5 L6
L7
L8
L9
L10 L11
53
microgrids as a result of reduction in curtailed electricity and heat demands. For instance, in the fourth
stage of the reinforcement, the electricity distribution lines L9 and L11 are reinforced. With this preventive
reinforcement, the operation cost is reduced by 51.5% compared to that of Case 4 and reached to $8,424. In
addition, the resilience index is increased to 0.6627. Table 3-4 shows the operation cost, reinforcement
cost, disrupted components and respective resilience index at each stage of reinforcement.
It is assumed that the encryption cost and consequently the disruption cost for the interrupted components
will double at each stage as a result of the applied reinforcements. Figure 4-8 illustrates the reduction in the
operation cost of the disrupted microgrid achieved by improving the encryption strategy at each stage of the
preventive reinforcement. By comparing Case 1 with Case 4, it is observed that with the increase in
disruption cost of the natural gas and electricity distribution network components, fewer islands are formed
within the microgrid and the disruptions would be on the more expensive components which have not been
reinforced earlier such as electricity generators in each island. It is also shown that the electricity and heat
demands with higher VOLL are the primary targets for disruption, and consequently, the electricity
distribution and natural gas pipelines supplying the energy hubs serving high priority demands are the first
options for preventive reinforcements. By adding the operation cost and the reinforcement cost, the total
cost at each stage of reinforcement is procured, which is shown in Figure 4-8. As shown in this figure, the
minimum reinforcement and operation cost is achieved at sixth stage of reinforcement. At this stage, the
total cost is $23,090. Hence, deploying the preventive reinforcement beyond this stage increases the
investment cost on improving the encryption strategy beyond the increase in the operation cost as a result
of disruption.
54
Table 3-4 Outcomes of preventive reinforcement procedure
Stage
Operation
cost
($)
Resilience
Index
(r)
Disrupted
Component
Total
Encryptio
n Cost
($)
0 26,275 0.2715 P4, L2, L3,
L4, L7 9,344
1 21,860 0.3385 L1, L2, L6,
L9, L10,L11 11,136
2 12,777 0.5331 L1,L8
L10,L11 12,298
3 12,770 0.5333 P1,L4,L5 14,720
4 8,424 0.6627 L9, L11 16,512
5 6,753 0.7205 G3, L3, L7 18,048
6 3,122 0.8639 L8, L10 19,968
7 1,924 0.9172 L1, P2 21,504
8 753 0.9725 P3 23,296
9 497 0.9851 P4 24,064
10 322 0.9937 P5 25,600
11 307 0.9945 P1 26,368
12 307 0.9945 P2 27,904
13 307 0.9945 P2 28,672
14 211 0.9993 G1 29,440
15 210 0.9993 G2 31,232
16 200 0.9998 G3 33,024
17 195 1 - 33,920
As shown in Table 3-4, after 17 stages of reinforcement, the microgrid is totally resilient to disruption
attempts with limited resources ($20,000 in this case study). More resources for disruptions will result in
more reinforcement steps within the multiple energy carrier microgrid. Figure 4-8 shows that as the
reinforcement cost increases, the operation cost of the microgrid which reflects the lost electricity and heat
demand, decreases.
55
Figure 3-8 Operation and reinforcement costs of multi-energy carrier microgrid.
3.5 Summary
This chapter introduced a methodology to analyze the resilience of microgrids with multiple energy
carrier networks exposed to interruptions in electricity and natural gas distribution networks. The
vulnerability of multiple energy carrier microgrids is addressed by introducing energy hubs which
emphasizes the interdependence of electricity and natural gas energy infrastructures. It is shown that
disruption in an energy network within the microgrid has adverse effects on other interdependent energy
networks serving electricity and heat demands. It is observed that in order to maximize the adverse effects
of disruptions the attacker inclines to create isolated networks within the microgrid as shown in Cases 1, 2,
and 4. The proposed methodology suggests the reinforcement strategies for microgrid operators and
investors by highlighting the vulnerable components and procuring the disruption outcomes as a result of
component failures considering the interdependence of multiple energy infrastructures within the
microgrid. The recommended preventive reinforcement warrants the resilient operation of multiple energy
carrier microgrids considering limited reinforcement budget and the desired level of resilience. This chapter
is focused on proposing a methodology to identify the vulnerable components and ensure the resilient
operation of coordinated electricity and natural gas infrastructures considering multiple disruptions within
0
5000
10000
15000
20000
25000
30000
35000
40000
10
100
1,000
10,000
100,000
0 2 4 6 8 10 12 14 16
Op
erat
ion
co
st (
$)
Iterations
Operation Cost
Reinforcement Cost
Total Cost
To
tal/
Rei
nfo
rcem
entC
ost
($)
56
the microgrid. Although the diverse load scenarios with different values of lost load in multiple periods
would result in a different preventive reinforcement plans for the operation period, the preventive
reinforcement procedure remains the same. Defining the multi-period problem requires a detailed definition
of the attacks addressing the length of interruption for each attack and the restoration time for the
components if exists. This problem could be addressed as the future work for this chapter.
57
Chapter 4
4A HIERARCHICAL ELECTRICITY MARKET STRUCTURE FOR THE SMART GRID PARADIGM
This chapter is organized as follows: Section 4.1 describes the problem formulation. Section 4.2
introduces the proposed multi-level dynamic game with complete information. Section 4.3 presents a case
study, and section 4.4 presents the summary.
4.1 Problem Formulation
A method for analyzing the competition among GENCOs in the WEM is described in [Li05]. The
competition is modeled as a bi-level optimization problem in which the upper-level problem represents the
profit maximization of the GENCOs and the lower-level problem represents the WEM clearing process. A
framework for calculating multi-participant Nash equilibria in a transmission-constrained electricity market
is presented in [Kho14], where the non-cooperative complete information game is formulated for market
participants with discrete bidding strategies. In this section, the profit maximization problem for each
market participant is presented as a mathematical problem with equilibrium constraints (MPEC) which is
formulated as a bi-level optimization problem. The upper-level problem addresses the profit maximization
while the lower-level problem is the operation cost minimization. The non-cooperative complete
information game for the WEM and the DNEM is represented as an equilibrium problem with equilibrium
constraints (EPEC). An EPEC is formed by several MPECs, in which the lower-level problems are the
same for all MPECs. The EPEC renders a generalized Nash equilibrium problem (GNEP) as the strategy
taken by each market participant is dependent on the decisions made by other participants in the WEM and
the DNEM [Gab12]. Hence the proposed hierarchical market is composed of multiple EPECs in which the
equilibrium at one EPEC is dependent on the equilibrium of the other interdependent EPECs.
58
4.1.1 GENCO’s objective
GENCO j maximizes the payoff function which is shown in (4.1), subjected to the constraints on
the bidding vector as shown in (4.2). The payoff function represents the revenue of GENCO, which is the
first term in (4.1) as well as the generation cost, which is the second term in (4.1). GENCO maximizes the
payoff by proposing the bidding vector, which is formulated in (4.3).
2
,
P ( )maxo
j j j
j o j j j j j jP o B
R LMP P P
ψ (4.1)
j j j ψ ψ ψ (4.2)
2 ;j j j je j je je
P ρ ψ P P (4.3)
4.1.2 Microgrid’s objective
Microgrid m maximizes the payoff function (4.4), subjected to the limitations on the bidding vector
as shown in (4.5). Here, the first term in the payoff function is the revenue of the microgrid and the second
term is the generation cost. The generation dispatch of microgrids is provided by curtailing the IL and
dispatching the DERs. The microgrid may adopt several policies to serve its demand. One policy is to serve
local demand and bid the excess generation in the DNEM while the other is to bid on its total generation
capacity in the DNEM and serve the local demand economically from the distribution network. In both
cases, the bidding vector for generation in microgrid m is shown in (4.6).
20
,maxm m
m m m m m m mP
R P P P ψ
(4.4)
m m m ψ ψ ψ (4.5)
2 ;m m m me m me me
P ρ ψ P P (4.6)
59
Here 0 is the Lagrangian multiplier associated with the generation and load balance constraint
within the distribution network.
4.1.3 LA’s objective
Each LA participates in the WEM and the DNEM. The LA acts as a broker playing two different
roles: a) buy electricity from the WEM and sell the same volume to the DNEM, if the LMP on the LA’s
bus in the WEM is lower than that in the DNEM. b) purchase electricity from the DNEM and sell the same
volume to the WEM, if the MCP in the DNEM is lower than the LMP on the LA’s bus in the WEM.
Considering role “a”, LA l maximizes the payoff function which is shown in (4.7), subjected to the
bidding vector constraint (4.8). Here LMP is the marginal price of electricity on the LA’s bus in the WEM
which is determined as the cost of providing electricity to the distribution network. The LA’s bidding
vector in the DNEM is given in (4.9).
0,
max [( ) ]imp o
l l
impl o l
P o B
R LMP P
lψ
(4.7)
l l l ψ ψ ψ (4.8)
ol
l l oo B
LMP
ρ ψ (4.9)
exp
exp0
,max
ol l l
l o lP o B
R LMP P
ψ
(4.10)
l l l ψ ψ ψ (4.11)
0l l ρ ψ (4.12)
Considering role “b”, the LA l maximizes the payoff function which is shown in (4.10), subjected
to the constraints on the bidding vector (4.11). The market clearing price ( 0 ) in the DNEM is the marginal
cost of providing electricity to the WEM. The bidding vector of LA l in the WEM is shown in (4.12).
60
4.1.4 The WEM clearing process
In the WEM, the ISO minimizes the operational cost subjected to generation and transmission
network constraints. The WEM clearing problem is formulated as (4.13) – (4.17).
expmin j j l l
j l
P P ρ ρ (4.13)
Subjected to: exp
0impD
j ol lj l o l
P P P P (4.14)
exp( . . ) . . , ,
o o oj l l
impo o o D ob b j b b o b b b bl l
o B o B o B
F SF P SF P SF P SF P F b
(4.15)
j j jP P P j (4.16)
exp exp expl l l
P P P l (4.17)
0 ( )T
oo b b b
b
LMP SF (4.18)
The objective function which is shown in (4.13), is the operation cost of the system which
includes the awarded dispatch of market participants multiplied by their bids. The load balance constraint is
shown in (4.14). The transmission line capacity constraint is shown in (4.15). The generation capacity
constraint is shown in (4.16), and the limitation on the electricity exported by each LA is shown in (4.17).
As shown in (4.18), the Lagrangian multipliers associated with the load balance and transmission line
capacity constraints are employed to calculate the LMPs on each bus in the WEM.
4.1.5 The DNEM clearing process
The DNO minimizes the operational cost of distribution network as shown in (4.19)-(4.22).
minimp
m m l lm l
P P ρ ρ (4.19)
Subjected to: exp
0imp D
m l lm l l
P P P P (4.20)
61
m m mP P P m (4.21)
imp imp impl l l
P P P l (4.22)
The load balance in the distribution network is shown in (4.20). The constraints on the awarded
generation for microgrids and the imported power by the LA are shown in (4.21) and (4.22) respectively.
Here, 0 is the Lagrangian multiplier for the generation and load balance constraint in the distribution
network as it is assumed that the electricity delivery is not limited by the line congestions.
4.2 Multi-level complete information game
In case of competition with complete information, each market participant recognizes the
opponents’ payoff functions and bidding strategies. Hence the WEM clearing problem which is shown in
(4.13)-(4.17), and the DNEM clearing problem which is shown in (4.19)-(4.22), are linear programming
problems (LP) that are re-written in general form as (4.23)-(4.25) and (4.26)-(4.28) respectively.
min Tc x (4.23)
Subjected to: A x = b π (4.24)
x x x θ (4.25)
min Tc x (4.26)
Subjected to: Ax = b π (4.27)
x x x θ (4.28)
The bi-level problem for GENCO j is composed of the upper-level problem i.e. the payoff
maximization, which is shown in (4.1)-(4.2); and the lower-level problem, which is shown in (4.23)-(4.25).
Accordingly, the bi-level problem for microgrid m is composed of the upper-level problem, which is shown
(4.4)-(4.5); and the lower-level problem, which is shown in (4.26)-(4.28).
62
Since the LA l participates in multiple markets, if the LA plays role “a”, the upper-level problem
is the payoff maximization problem which is shown in (4.7)-(4.8) and the lower-level problem is shown in
(4.26)-(4.28). If the LA plays role “b”, the upper-level problem is shown in (4.10)-(4.11) and the lower-
level problem is shown in (4.23)-(4.25). In the three types of the bi-level problems presented, the upper-
level problems are the payoff maximization of the GENCOs, microgrids, and LAs while the lower-level
problems represent the operation cost minimization by the ISO or the DNO, subjected to the prevailing
constraints. The proposed bi-level problems are solved for the market participants in the WEM and the
DNEM simultaneously to determine the bidding strategies of market participants in the respective markets.
In this paradigm, microgrids and LAs participate in the DNEMs while GENCOs and LAs bid in the WEM.
Each LA which links the upper level market (WEM) and the lower level market (DNEM), determines its
bidding strategy and role based on the outcomes of each electricity market. The interaction between the
market participants in the WEM and the DNEM continues until none of the market participants in both
markets would update their bidding strategies. The synergy between the market participants in the WEM
and the DNEM is presented as Cournot game while the interaction between the WEM and the DNEMs is
presented as a dynamic game. Here microgrids, GENCOs, and LAs update their bidding strategies, based
on the sensitivity of the payoff function to the bidding strategy chosen. Considering the WEM, the bi-level
linear problems are converted into single level nonlinear problems using the Karush-Kuhn-Tucker (KKT)
optimality conditions for the lower-level problems. Assuming that the lower bound of inequality (4.25) is
zero, the KKT optimality conditions for (4.23)-(4.25) are presented in (4.29)-(4.34). Here ( )Diag X x ,
( )Diag Φ φ , ( )Diag Τ τ and ( )Diag Θ θ .
ΤΑ π + τ -θ = c (4.29)
Α x = b (4.30)
x +φ = x (4.31)
X τ = 0 (4.32)
Φ θ = 0 (4.33)
63
(x ,τ ,θ ,φ ) 0 (4.34)
By differentiating (4.29)-(4.34) with respect to the GENCOs’ and LAs’ bidding vector
components, the equality (4.35) is obtained.
Here the coefficient on the left hand side and the term on the right hand side of the system of
equations (4.35) are Η and d c respectively. In (4.35), H is a low rank sparse matrix, and once the least-
square technique [Woo96] is employed to solve this system of nonlinear equations, the sensitivity function
vector denoted as [ ]Td z π ψ x ψ τ ψ θ ψ is given as (4.36).
T
π
ψc
xA 0 I -Iψ
ψ0 A 0 00
τ0 Τ X 00
ψ0 -Θ 0 Φ0
θ
ψ
(4.35)
1( )T Td d z H H H c (4.36)
In the proposed iterative approach, the payoff of each market participant is maximized by
calculating d z from (4.36) iteratively. The sensitivity functions calculated in (4.36) are employed in (4.37),
(4.38), (4.41), and (4.42) to acquire the sensitivity of the payoff of each market participant with respect to
its bidding strategy at each iteration. The sensitivity of the payoff of GENCO j with respect to the bidding
strategy chosen is shown in (4.37). Other components in (4.37) were calculated in (4.38)-(4.40).
k k k k kj j j j oo
jk k k k kj j j o j
R R P R LMPj B
P LMP
(4.37)
, ,0 .
k kk ko oo b b
b jk k k kbj j j j
LMPSF j B
(4.38)
(2 )
kj k k o
o j j j jkj
RLMP P j B
P
(4.39)
64
kj k o
j jko
RP j B
LMP
(4.40)
Similar formulation is applied for microgrid m in the DNEM, with respective variables and indices.
For LA l with bidding strategy kl , the sensitivity of the payoff with respect to the bidding strategy
chosen, is shown in (4.41). The components which are shown in (4.41), are calculated in (4.42)-(4.44).
exp, exp, exp, exp,
exp,
k k k k kol l l l olk k k k k
l l o ll
R R P R LMPl B
LMPP
(4.41)
, ,0
k kk ko oo b b
b lk k k kbl l l l
LMPSF l B
(4.42)
exp,
0exp,
kk k ol
o lkl
RLMP l B
P
(4.43)
exp,exp,
kk ol
llko
RP l B
LMP
(4.44)
Similar formulation is applied for LA l importing power to the distribution network. Here (4.45)
and (4.46) are employed for updating the bidding strategies of GENCOs and LAs in the WEM,
respectively, and is the constant for each step. Similar formulation is applied for updating the bidding
strategies of microgrids and LAs importing power to the distribution network.
1kjk k
j j kj
R
(4.45)
exp,1
kk k l
l l kl
R
(4.46)
The presented game is considered as a dynamic game as the LAs observe the actions of other
market participants in one market before updating their bidding strategy in the other. Here LAs consider the
decision made by GENCOs once bidding in the DNEM. Similarly, LAs observe the decision made by
65
microgrids once bidding in the WEM. The bidding vector of LAs in each market is determined based on the
awarded dispatch and the price of electricity at the LAs’ bus in the other market. The observability
indicator v is a binary parameter, which indicates that the LAs observed the actions of participants in the
other market at each iteration. The bidding strategy of market participants update iteratively until the
convergence criteria are satisfied and the generalized Nash equilibrium (GNE) is established. The presented
game may have one, multiple or no GNE. However, the market regulations will limit the bidding strategy
of market participants and improve convergence to a state in which the market participants are either
unwilling to unilaterally update their bidding strategy or they cannot do so. As illustrated in Figure 4-1, the
steps (a)-(n) are taken to procure the GNE of the proposed dynamic game. While multiple LAs could
participate in the WEM, for the sake of simplicity, only one LA (LA l ) is considered in this algorithm. The
following approach can be extended to address multiple LAs in multiple DNEMs.
a) Set the initial value of power import/export of the LA to zero with respective arbitrary large initial
marginal costs and initiate
0 0 0 0j m l l ψ ,ψ ,ψ ,ψ for each generator, microgrid, and LA; then go to “b”.
b) Set the iteration indices 1k and LA’s observability indicator 0v ; then go to “c”.
c) If the inequality (4.47) holds LA imports electricity to DNEM; then go to “d”. Otherwise go to “e”.
, 1 exp, 1imp k v k vl l
P P
(4.47)
d) Set the parameters using (4.48)-(4.50) for the DNEM while the LA’s role is to import electricity to this
market; then go to “f”.
1k k vo oLMP LMP (4.48)
, , 1 exp, 1imp k imp k v k vl l l
P P P
(4.49)
exp,0
kl
P (4.50)
e) Set the parameters using (4.48), (4.51), and (4.52) for DNEM while LA’s role is to export electricity
from this market; then go to “f”.
66
, , 1imp k imp k vl l
P P
(4.51)
exp, exp, 1k k vl l
P P
(4.52)
f) DNO clears DNEM; then go to “g”.
g) The bidding strategy of microgrids and LA l within the distribution network are updated using the
sensitivity functions of their payoff with respect to their bidding strategy; then go to “h”.
h) If the inequality (4.53) holds, LA exports electricity to the WEM; then go to “i”. Otherwise go to “j”.
, exp, 1imp k k vl l
P P
(4.53)
Figure 4-1 Competition in the hierarchical electricity market
LA export electricity
from distribution
market (e)
LA export electricity(h)?
Distribution Network
Market of LA l
Choose initial bidding strategies for all GENCOs
and LAs and microgrids (a)
Initial iteraction indices and indicators k=1, v=0 (b)
DNO clears the distribution network market (f)
ISO clears the wholesale electricity market(k)
Microgrids/LA l update bidding strategies (g)
Y
LA export electricity
to wholesale market
(j)
LA import electricity
from wholesale
market (i)
N
GENCOs/LA l update bidding strategies (l)
LA imported electricity(c)?N
Y LA import electricity
to distribution market
(d)
Toggle vObserved(m)?
Y
k=k+1
N
Nash Equilibiruim
(Optimal bidding Strategy for each market participant)
Converged(n)?
Y
N
67
i) Set the parameters using (4.54)-(4.56) for the WEM while LA’s role is to export electricity to the WEM;
then go to “k”.
10 0k k v (4.54)
exp, exp, 1 ,k k v imp kl l l
P P P
(4.55)
,0
imp kl
P (4.56)
j) Set the parameters using (4.54) and (4.57) for the WEM while LA’s role is to import electricity from the
WEM.
exp, exp, 1k k vl l
P P
(4.57)
k) ISO clears the WEM; then go to “l”.
l) The bidding strategy of the GENCOs and LA l within the WEM are updated using the sensitivity
functions of their payoff with respect to their bidding strategy.
m) Check if the observability indicator is equal to 1.0, go to “n”. Otherwise, toggle the observability
indicator go to “c”.
n) If the conditions (4.58),(4.59) hold for all market participants in the WEM and DNEM, then GNE is
established; otherwise, set 1k k , set the observability indicator to zero, and go to “c”.
1(.) (.)k k (4.58)
1(.) (.)k k (4.59)
Assuming the existence of an equilibrium, convergence to different local equilibriums depends on
the initial bidding strategies. As the feasible region for each market participant is non-convex, the
equilibrium among MPECs represent a local optimum solution. In [Rui12], the MPECs are rewritten as a
mathematical problem with primal dual constraints (MPPDC). Considering fixed dual variable associated
68
with the strong duality constraint of the MPPDC, the EPEC is further formulated as a parameterized mixed-
integer linear problem (MILP). The solution of the MILP is a local Nash equilibrium or a saddle point
depending on the choice of the fixed dual variable. In [She84], the existence of a unique equilibrium under
certain assumptions is proved, while the existence and uniqueness of an equilibrium is proved in [DeM09]
by relaxing some constraints. The alternative method to the sensitivity function employed in this chapter, is
to reformulate the problem in to nonlinear complementarity problem (NCP) [Ley10], [Fle04].
4.3 Case study
In order to evaluate the effectiveness of the proposed approach, a WEM with two generators, and
one DNEM including two microgrids and one LA are considered. Figure 4-2 shows the proposed network
configuration. Here, the DNEM and the WEM have the total demand of 75 MW and 260 MW respectively.
Table 4-1 shows the network data for both electricity markets. The coefficients of the generation
cost curve for GENCOs and microgrids are shown in Table 4-2. As shown in this table, the marginal cost of
electricity generation for the microgrids are higher than that for the GENCOs. The lower and upper limits
for the bidding strategy vector for market participants are 0.5 and 4.0 respectively. The lower limit for the
bidding strategy of LA is 1.0. The initial bidding strategy for LA is set to 1.0. The multiplier and the
threshold for convergence are set to be 1E-5 and 1E-4, respectively. The initial marginal cost of LA in
the WEM and DNEM is 1000$/MWh.
Table 4-1 Transmission and distribution network data
Line From
bus
To
bus
Reactance
(p.u)
Capacity limit
(MW)
L1 3 4 0.0098 160
L2 4 5 0.0105 250
L3 5 6 0.02 190
L4 6 3 0.01 130
L5 1 3 0.005 100
L6 2 3 0.008 90
Table 4-2 Generation cost curve of microgrids and GENCOs
Market
Participant
Min.
capacity
(MW)
Max.
capacity
(MW)
($/MW2)
($/MW)
($)
MG1 0 60 0.125 25 0
MG2 0 60 0.250 20 0
G1 0 210 0.073 6.926 0
G2 0 210 0.066 7.352 0
69
Figure 4-2 Power network topology
The following three cases are presented:
Case 1) higher marginal cost of electricity in microgrids with no congestion in the wholesale electricity
market.
Case 2) higher marginal cost of electricity in microgrids with congestion in the wholesale electricity
market;
Case 3) lower marginal cost of electricity in microgrids with no congestion in the wholesale electricity
market.
4.3.1 Case 1) higher marginal cost of electricity in microgrids with no congestion in the wholesale
electricity market
In this case, the microgrids provide higher marginal cost for electricity compared to the GENCOs as shown
in Table 4-2. The bidding strategies and the awarded dispatches for each market participant are listed in
Table 4-3. Since there is no congestion in the WEM, the MCP is 80 $/MWh, while the MCP in the DNEM
is 140 $/MWh. In this case, the lower price of electricity in the WEM provides incentives for the LA to
import electricity from this market to the DNEM. The payoff of MG1, MG2, LA, G1 and G2 are $4,400,
MG1 MG2
2
G2
5
3
6
50MW 25 MW
100 MW
4
G1
90 MW
70 MW
1
Distribution Network
Electricity Market
Wholesale
Electricity Market
70
$2,300, $900, $8,800, and $8,604 respectively. Hence the LA spent $1,200 to purchase 15MW electricity
and earned $2,100 by selling the same volume in the DNEM. The exchanged power, the LMP, and the
bidding strategies of the market participants in the WEM and the DNEM are illustrated in Figure 4-3.
In the first iteration an arbitrary large number is assigned as the LMP at the LA’s bus in the WEM
(1000$/MWh) and LA bids into the DNEM based on the assigned LMP at the WEM. The awarded dispatch
imported to DNEM and the MCP at the DNEM is determined. As shown in Figure 4-3a, in iterations 2 to
36, the bidding strategy of G1 is greater than G2, which in turn is greater than the bidding strategy of the
LA in the WEM. Similarly, as shown in Figure 4-3b, the bidding strategy of MG1 is greater than that of
MG2, which in turn is greater than the bidding strategy of the LA in the DNEM. As shown in Figures 4c
and 4d, the LA is awarded 20MW in the DNEM and the MCP in the DNEM is significantly larger than the
LMP at LA’s bus in the WEM. At iteration 95, the LMP at the LA’s bus in the WEM becomes closer to the
MCP of DNEM which would lead to zero power import by the LA in the DNEM. As the demand in the
DNEM is inelastic to the MCP, MG1 bids lower than MG2 to increase the generation in the DNEM as
shown in Figure 4-3b. At iteration 171, the LMP at the LA’s bus in the WEM is larger than that in the
DNEM which inspires the LA to export electricity from the DNEM as illustrated in Figure 4-3a. However,
the response of the GENCOs to such decision will alter the exported dispatch of the LA in the WEM. At
iteration 189, G2 reaches the upper limit of the bidding strategy regulated by the market rules. In response,
the bidding strategy of G1 reaches 3.64, and the LMP of LA’s bus in the WEM will be 80 $/MWh as
shown in Figures 4a and 4d. Ultimately as shown in Figure 4-4, the LA is awarded 15 MW in the DNEM at
iteration 204, as the MCP at the DNEM is higher than the LMP of LA’s bus at the WEM. This situation
continues until MG2 reaches its highest possible bidding strategy (i.e. 4.0) limited by the market regulation.
As illustrated in Figure 4-3b, the bidding strategy of MG1 reaches 3.74, and the MCP of DNEM becomes
140 $/MWh.
71
(a) Bidding strategy of market participants in the WEM
(b) Bidding strategy of market participants in the DNEM
(c) Exchanged power for the LA (negative/positive values represent imported power
to/exported power from the DNEM)
(d) LMP of LA’s bus in the WEM and MCP of DNEM
Figure 4-3 Bidding strategies of market participants, exchanged power, and energy price in Case
0
1
2
3
4
1 51 101 151 201 251 301B
idd
ing S
trat
egy
Iteration
LAG1G2
0
1
2
3
4
1 51 101 151 201 251 301 351 401 451 501
Bid
din
g S
trat
egy
Iteration
MG1
MG2
LA
-40
-20
0
20
40
0 50 100 150 200 250
Exch
nag
ed
Po
wer
[M
W]
Iteration
-10
40
90
140
1 51 101 151 201 251 301 351 401 451 501
LM
P [
$/M
Wh]
Iteration
MCP of DNEM LMP of LA in WEM
72
Table 4-3 Bidding strategy and awarded dispatch in Case 1 MG1 MG2 LA G1 G2
Bidding Strategy 3.74 4 1.05
(imports power) 3.64 4
Awarded Dispatch
(MW) 40 20 15 140 135
Figure 4-4 Market clearing results in Case 1
4.3.2 Case 2) higher marginal cost of electricity in microgrids with congestion in the wholesale
electricity market
In this case, the impact of transmission line congestion is considered in two different cases.
4.3.2..1 Congestion in Line 1
In this case, the capacity limit of line L1 is set to 70 MW. Because of the congestion on L1, the
LMP on bus 3 in the WEM is increased to 841.20 $/MWh while the MCP in the DNEM is 180 $/MWh as
shown in Figure 4-5. This would provide incentives for the LA to export 27.25MW from the DNEM to the
[40 MW]
MG1
[20 MW]
MG2
2
[135 MW]
G2
5
3
6
L4
L1
50MW 25 MW
100 MW
4
L2
L3
[140 MW]
G1
90 MW
70 MW
1
Distribution Network
Electricity Market
Wholesale Electricity
Market
[MCP: 80 $/MWh]
[MCP: 140 $/MWh]
[15 M
W]
1 3.74 2 4
1.05l
1 3.64
2 4
73
WEM. As listed in Table 4-4, G2 is only awarded 22.75 MW with the LMP equal to 48 $/MWh, while G1,
is awarded its maximum generation capacity of 210 MW with the LMP equal to 253.78 $/MWh on bus 4.
In this case MG1’s strategy is to sell more electricity with lower price while MG2’s strategy is to sell less
electricity with higher price. The payoff of MG1, MG2, LA, G1 and G2 are $8,850, $6,313, $18,802,
$48,620, and $886 respectively. Here, the payoff of G2 decreased dramatically compared to that in Case 1,
as a result of congestion while the payoffs of other market participants especially G1 and LA were
increased.
Table 4-4 Bidding strategy and awarded dispatch in case 2 (L1 is congested) MG1 MG2 LA G1 G2
Bidding Strategy 1 4 4
(exports power) 4 4
Awarded Dispatch (MW) 60 42.25 27.25 210 22.75
Figure 4-5 Market clearing result in Case 2.1
4.3.2..2 Congestion in Line 4
In this case, the capacity limit of line L4 is set to 30MW. As a result of the congestion on line L4,
the LMP on bus 3 in the WEM is 72.53 $/MWh which is lower than that in Case 1. The awarded dispatch
[60 MW]
MG1
[42.25 MW]
MG2
2
[22.75 MW]
G2
5
3
6
L4
L1< 70 MW
50MW 25 MW
100 MW
4
L2
L3
[140 MW]
G1
90 MW
70 MW
1
Distribution Network
Electricity Market
Wholesale Electricity
Market
[MCP: 180 $/MWh]
[27
.25
MW
]
1 1 2 4
4l
1 3.64
2 4
[LMP: 841.2 $/MWh]
[LMP: 48 $/MWh]
[LMP: 253.8 $/MWh]
74
for the LA in the DNEM is 15 MW. The imported electricity to the DNEM mitigates the congestion on line
L4 in the transmission network, and the LA has the incentive to increase the electricity import to the
DNEM to reduce the LMP at bus 3 in the WEM.
As shown in Figure 4-6, the MCP in the DNEM is 130 $/MWh, which is lower than that in Case 1. As
shown in Table 4-5, G2 is awarded 117.52 MW with the LMP equal to 80 $/MWh while G1 is awarded
157.48 MW with the LMP equal to 88 $/MWh as a result of congestion on line L4. MG1 and MG2 were
awarded 20 MW and 40 MW in the DNEM respectively. In this case, MG2 proposes lower bid resulting in
the increase in its awarded generation and the decrease in the imported power from the WEM. The payoff
of MG1, MG2, LA, G1 and G2 are $2,050, $4,000, $862, $10,956, and $7,626 respectively. In this case,
the payoffs of LA, MG1, and G2 decreased compared to those in Case 1, while MG2 and G1 had much
higher payoffs as a result of congestion on line 4.
Table 4-5 Bidding strategy and awarded dispatch in case 2 (L4 is congested)
MG1 MG2 LA G1 G2
Bidding Strategy 4 2.89 1.02
(imports power) 4 4
Awarded
Dispatch (MW) 20 40 15 157.48 117.52
Figure 4-6 Market clearing results in Case 2.2
[60 MW]
MG1
[42.25 MW]
MG2
2
[22.75 MW]
G2
5
3
6
L4
L1< 70 MW
50MW 25 MW
100 MW
4
L2
L3
[140 MW]
G1
90 MW
70 MW
1
Distribution Network
Electricity Market
Wholesale Electricity
Market
[MCP: 180 $/MWh]
[27.2
5 M
W]
[LMP: 841.2 $/MWh]
[LMP: 48 $/MWh]
[LMP: 253.8 $/MWh]
75
4.3.3 Case 3) lower marginal cost of electricity in microgrids with no congestion in the wholesale
electricity market
In this case, the generation costs for microgrids are reduced to 20% of those shown in Table 4-2,
which is lower than the marginal costs of GENCOs in the WEM. The bidding strategy and the awarded
dispatch for each market participant are listed in Table 4-6. Here, the LMP at all buses in the WEM is 80
$/MWh, while the MCP in the DNEM is 36 $/MWh. The payoffs of MG1, MG2, LA, G1 and G2 are
$1,770, $1,740, $1,980, $8,800, and $5,077 respectively. Here the LMP at LA’s bus in the WEM is higher
than the MCP in the DNEM, hence, the LA exports power from the DNEM. Compared to Case 1, G1 has
the same profit, the profit of G2 is decreased, and the profit of LA is increased. Since the LMP at LA’s bus
in the WEM is higher than that in the DNEM, the LA exports electricity to the WEM.
Table 4-6 Bidding strategy and awarded dispatch in case 3 MG1 MG2 LA G1 G2
Bidding Strategy 1.75 4 1
(exports power) 3.64 4
Awarded Dispatch
(MW) 60 60 45 140 75
Once the market participants bid on their marginal prices of electricity, the payoff of MG1, MG2,
LA, G1 and G2 are $697, $667, $84, $400, and $577 respectively and the LMP at all buses in the WEM is
20 $/MWh and the MCP in the DNEM is 18.12$/MWh. In this case, the LA export 45 MW to the WEM.
The total payoff of all market participants is decreased by $15,046 in this case compared to the case in
which the market participants bid strategically.
Figure 4-8 shows the bidding strategies of the market participants in the WEM and the DNEM, the
LMP of LA’s bus in the WEM, as well as the MCP of DNEM. Similar to Case 1, in the first iteration an
arbitrary large number is assigned as the LMP at the LA’s bus in the WEM (1000$/MWh) and LA bids in
the DNEM based on the assigned LMP at WEM. In this case, the awarded dispatch imported to the DNEM
is zero and the MCP at DNEM is determined. As shown in Figure 4-8a, in all of the iterations, the bidding
strategy of G2 is greater than that of G1, which in turn is greater than the bidding strategy of LA in the
WEM. As shown in Figure 4-8b, in iterations 2 to 353, the bidding strategy of MG1 is close to that of MG2
and the bidding strategies of MG1 and MG2 are greater than the bidding strategy of the LA in the DNEM.
76
As shown in Figures 4-8c, the LMP at LA’s bus in the WEM is significantly larger than the MCP in the
DNEM, hence the LA is awarded 45MW in the WEM. At iteration 283, G2 reaches the upper limit for the
bidding strategy regulated by the market. In response, the bidding strategy of G1 reaches 3.64, and the
LMP of LA’s bus in the WEM will be 80 $/MWh as shown in Figures 4-8a and 4-8c. The bidding strategy
of MG1 reaches 1.75 in iteration 363. This situation continues until MG2 reaches its highest possible
bidding strategy (i.e. 4.0) limited by the market regulation, and the MCP of DNEM becomes 36 $/MWh, as
illustrated in Figure 4-8b.
Figure 4-7 Market clearing results in Case 3
[60 MW]
MG1
[60 MW]
MG2
2
[75 MW]
G2
5
3
6
L4
L1
50MW 25 MW
100 MW
4
L2
L3
[140 MW]
G1
90 MW
70 MW
1
Distribution Network
Electricity Market
Wholesale Electricity
Market
[MCP: 36 $/MWh]
[45 M
W]
[MCP: 80 $/MWh]
77
(a)
(b)
Figure 4-8 Bidding strategies of market participants and energy prices in Case 3 (a) Bidding
strategy of market participants in WEM. (b) Bidding strategy of market participants in DNEM.
(c) LMP of LA bus in WEM and MCP of DNEM
0
1
2
3
4
1 51 101 151 201 251 301
Bid
din
g S
trat
egy
Iteration
LAG1G2
0
1
2
3
4
1 201 401 601 801
Bid
din
g S
trat
egy
Iteration
MG1
MG2
LA
0
20
40
60
80
100
1 201 401 601 801
LM
P [
$/M
Wh]
Iteration
MCP of DNEM LMP of LA in WEM
4.4 Summary
Microgrids are the building blocks of active distribution networks in the smart grid paradigm
which provide energy and ancillary services in distribution and wholesale electricity market. In this
chapter, the hierarchical market structure for the smart grid paradigm which is composed of the WEM and
the DNEM is proposed. Here the LA represents as the middle agent which participates in the distribution
network and the WEMs competing with microgrids and GENCOs. The competition among market
participants in both electricity markets with complete information is presented as a dynamic game. The
bidding strategy chosen by each market participant is procured using bi-level linear programming problem
with the upper-level problem representing the market participant’s payoff maximization and the lower-level
problem minimizing the operation cost of the network. The bi-level problems are solved by developing
78
sensitivity functions for the market participants’ payoff with respect to their bidding strategies. The result
shows the impact of LA’s bidding strategy on the bidding strategy and payoffs of other market participants
in the WEM and the DNEMs. While the non-cooperative dynamic game with complete information is
proposed in this chapter, the proposed approach can be extended to address the non-cooperative dynamic
game with incomplete information.
79
Chapter 5
5RISK AVERSE GENERATION MAINTENANCE SCHEDULING WITH MICROGRID
AGGREGATORS
This chapter presents a risk-averse formulation for generation maintenance scheduling using microgrid
aggregators in the bulk power system. Microgrid aggregators capture the characteristics of aggregated
distributed generation units and demand served by a large number of microgrids. The system operators
leverage the proposed framework in order to determine the most effective long-term generation
maintenance schedule while ensuring the secure and economic operation of power systems through
balancing the demand and supply. The system operator incorporates the characteristics of the generation
units for maintenance practices including the minimum and maximum periods in which the unit should be
on scheduled outage, the duration of the maintenance, and respective maintenance costs. Moreover, the
system operator determines the marginal cost of the generation units and local generation assets for
microgrid aggregators based on the submitted marginal costs (bids) for the short-term operation practices.
The individual characteristic of the local generation in the microgrids is inherently captured by the
aggregated generation bid, generation capacity, and respective uncertainties in such values. The presented
formulation is a two-stage model in which the first stage captures multiple maintenance practices in long-
term operation horizon by introducing the minimum and maximum time for the maintenance practices, and
in the second stage, the worst realization of uncertainties in the long-term maintenance scheduling is
revealed. The uncertainty in available generation resources of microgrids, the demand of the power system,
the marginal cost of the microgrids as well as contingencies in the transmission network are addressed and
further limited by the budget of uncertainty. The contributions of this chapter are as follows:
• The long-term generation maintenance scheduling of power system is formulated as a two-stage
robust optimization problem in which the first stage addresses the maintenance scheduling of
the generation units, while the second stage captures the operation decisions once the worst
realization of the uncertainties is revealed.
• A new formulation for long-term maintenance scheduling is presented that addresses multiple
80
maintenance practices for generation units considering the minimum and maximum periods for
the availability of the units prior to maintenance, as well as the required number of periods for
the maintenance practices.
• The presented formulation determines the worst realization of uncertainties in the marginal cost
of the microgrid generation, the capacity of the local generation assets in microgrids, electricity
demand, and the availability of the transmission network components considering a certain
budget for the uncertainties in the power system. Furthermore, the impact of the budget of
uncertainty on the operation and maintenance costs is evaluated.
• The effect of the aggregated microgrids on the long-term generation maintenance scheduling is
addressed by performing sensitivity analysis on the capacity of local generation and the
marginal cost of electricity in microgrids.
The rest of the chapter is organized as follows, the problem formulation and solution methodology are
described in Sections 5.1 and 5.2 respectively. A case study to show the effectiveness of the problem is
shown in Section 5.3. The summary is presented in Section 5.4.
5.1 Problem Formulation
The mathematical formulation for the generation maintenance scheduling is presented in (5.1)-(5.24). The
objective is to minimize the maintenance and the operation cost of the power system as shown in (5.1). The
first term in (5.1) represents the maintenance cost for all generation units in the power system and the
second term represents the operation cost of the system. The operation cost of the system is affected by
several uncertain variables in the operation horizon. These uncertain variables are the demand in power
system, the available local generation assets in microgrids that reduces the net demand realized by the
power system, the marginal cost of the generation units in microgrids, and the contingencies in the
transmission networks that changes the power flow pattern in the system and affects the operation cost. The
second term in (5.1) represents the operation cost minimization that is further maximized over the variables
in the uncertainty set. The operation cost of each generation unit is a quadratic cost function of the
generation dispatch, which is further linearized using a piece-wise linearization technique as shown in
(5.2)-(5.4) [Sha02]. The dispatch of the generation unit is limited by the maximum capacity of the unit as
81
shown in (5.5). The relationship between the binary variables representing the transition of the unit to
outage state for maintenance, the binary variables representing the transition from outage to available state,
and the state of the unit at each period is shown in (5.6). As shown in this constraint, once the unit goes on
maintenance from the available state, ,i tX will be 1 and the unit will change its state from available to
unavailable. Similarly, once the unit becomes available from an outage state in the maintenance period, ,i tY
will be 1 and the state of the unit ( ,i tI ) will change from 0 to 1. The number of periods in which the
generation unit was available at the beginning of each period is determined by (5.7) and (5.8). Here, if the
state of the generation unit is not transitioning from outage to available mode, the number of periods that a
generation unit was available at the beginning of each period is increased by one. The number of periods
the generation unit should be available is less than a maximum limit as shown in (5.9) and more than a
minimum limit as enforced by (5.10). Therefore, the period in which the unit should go on maintenance is
defined by constraints (5.9) and (5.10). Once the unit is recovered from the scheduled maintenance, the unit
will be available for the minimum number of periods and can go on maintenance for the second time
between the minimum and maximum periods assigned for the maintenance practices. The transition states
are mutually exclusive as shown in (5.11). The number of periods in which the unit is on maintenance is
determined by (5.12) and (5.13). The number of periods the unit should be on maintenance is enforced by
(5.14)-(5.15). Here, once the unit is recovered from the outage state to the available state, ,i tX is 0, and
,i tY is 1. Therefore, the number of periods the unit is on maintenance is enforced to be equal to iRHF . The
generated power at each bus is determined by (5.16) and the nodal power balance in the power system is
determined by (5.17). The power transmitted through the transmission line is dependent on the difference
between the voltage angles of the interconnected buses as shown in (5.18)-(5.19) if the transmission line is
available. Here, once the transmission line is disconnected because of contingency, the voltage angles of
the buses connected by the transmission line are relaxed. The power transmitted through the transmission
line is limited by the capacity of the transmission line as shown in (5.20). The capacity of the local
generation asset in microgrids connected to each node of the power system, is limited to a certain portion of
the demand at the corresponding node as shown in (5.21). The uncertainty in the marginal cost of the
generation units in microgrids, the uncertainty in the capacity of local generation assets in microgrids, and
82
the uncertainty in the demand of the power system are limited by (5.22), (5.23), and (5.24) respectively. As
shown in (5.22)-(5.24), the uncertain variables are identified by the nominal values and the uncertainty
interval. The available generation capacity of microgrid aggregators is dependent on the available energy
resources (e.g. renewable resources and fuel). Furthermore, the marginal cost of generation within
microgrids is dependent on several factors including the generation technology, the fuel cost, and the
inflation and interest rates. In order to capture the operation and maintenance costs, the investment costs,
the inflation and interest rates, and the labor cost associated with the electricity generation within the
microgrids, the Levelized Cost of Energy (LCOE) for the generated electricity within microgrids can
replace the marginal cost. As the number of microgrids is large and the generation technology utilized in
each microgrid could be different, capturing the marginal cost and generation capacity with respective
uncertainties for each microgrid is challenging and practically infeasible in this problem. Therefore, the
marginal cost and capacity of the local generation resources in microgrids with respective uncertainties are
aggregated and associated with the microgrid aggregators.
,, , , ,
, ,, , , ,
, , ,1 1 1 1 1
min max ming g d
i tl t b t b t b t
NT NG NT NG NB NMg m g m
i i t i t b t b tPu C P Pt i t i b m
MC X F C NW P
(5.1)
, ,1
NSs s
i t i i ts
F NW C P
1,: i t
(5.2)
, ,1
NSs
i t i ts
P P
2,: i t
(5.3)
,max,s s
i t iP P
1, ,: i t s
(5.4)
max, ,i t i i tP P I
2,: i t
(5.5)
, , , , 1i t i t i t i tY X I I (5.6)
, , 1 , 1 ,i t i t i t i i tHO HO I MHO Y (5.7)
, , 1 , 1 ,i t i t i t i i tHO HO I MHO Y (5.8)
83
, ,(1 )i t i i tHO MHO Y (5.9)
, ,i t i i tHO NHO X (5.10)
, , 1i t i tX Y (5.11)
, , 1 , 1 ,(1 )i t i t i t i i tHF HF I RHF X (5.12)
, , 1 , 1 ,(1 )i t i t i t i i tHF HF I RHF X (5.13)
, ,(1 )i t i i tHF RHF X (5.14)
, ,i t i i tHF RHF Y (5.15)
, , ,1
NG
i b i t b ti
A P P
3,: b t
(5.16)
,, , , ,,
NLg md
b t b t b l l tb tm l
P P P S f
4,: b t
(5.17)
, , ,(1 )NB
tl t l t b l b l
b
f M u S X
6,: l t
(5.18)
, , ,(1 )NB
tl t l t b l b l
b
f M u S X
7,: l t
(5.19)
max, ,l t l l tf f u
3 4, ,: ,l t l t
(5.20)
,, ,,
g m db t b tb t
m
P P
5,: b t
(5.21)
, , ,0 , , ,0 ,, , , , ,
,g m g m g m g m g mb t b t b t b t b t
C C C C C (5.22)
0 0, , , , ,,b t b t b t b t b t
(5.23)
84
,0 ,0, , ,, ,
,d d d d db t b t b tb t b t
P P P P P (5.24)
The presented problem is formulated as a two-stage robust optimization problem [Ber13] in which, the
first stage problem determines the first stage i.e. “here-and-now” variables including the scheduled
maintenance of the generation units in the power system, and the second stage problem will determine the
second stage i.e. “wait-and-see” variables once the uncertainties are revealed. These variables include the
generation dispatch of the available units, the flow of the transmission lines, the dispatch of local
generation assets, as well as the realization of the uncertainties including the marginal cost and capacity of
the microgrids’ generation, availability of the transmission lines, and the power system demand. The
uncertain variables determined at this stage belong to a polyhedral uncertainty set and the objective is to
minimize the sum of first-stage and second-stage costs considering a budget for the uncertainty that
represents the conservativeness of the decision maker and the size of uncertainty interval.
5.2 Solution Methodology
The column-and-constraint generation approach as a cutting plane procedure is proposed to solve the
presented two-stage robust optimization problem [Zen13]. This approach has superior computation
efficiency compared to the Benders decomposition [Ji16]. The problem presented in (5.1)-(5.24) is
presented in general form as (5.25)-(5.28) in which x is the first stage decision variable while y and u are
the second stage (i.e. recourse) decision variables. The objective function (5.1) is represented as (5.25) and
the set of constraints (5.26) represents all constraints with binary decision variables including the states and
the transition states of the units that were formulated in (5.5)-(5.15). The set of constraints shown by (5.27)
captures the second stage decision variables y such as the dispatch of generation units as well as the
realization of the uncertain variables u such as the demand of power system and the capacity of local
generation assets in microgrids. The feasibility set for first stage and second stage decision variables are
shown in (5.28). As enumerating all scenarios with different realization of u is practically challenging, the
solution approach leverages partial enumeration over a subset u considering the budget of uncertainty to
determine a valid relaxation to the original problem. The presented column-and-constraint generation
85
procedure further determines significant scenarios that contribute to the worst realization of the system
operation cost.
min maxminT T
x yu
c x b y
(5.25)
Subjected to: Ax d (5.26)
Fx Gy h Hu (5.27)
, x yx Ω y Ω (5.28)
The column-and-constraint generation procedure is implemented as a master problem and a sub-problem.
The algorithm for this procedure is shown as below:
a) Set iteration 0, ,LB UB with the determined realization of uncertainty *(0)u ,
b) Solve
,min
T ex y
c x (5.29)
Subjected to: ( )Te b y (5.30)
Ax d (5.31)
( ) *( ) Fx Gy h Hu (5.32)
, x yx Ω y Ω (5.33)
c) Set * *TLB e c x where *x and *e are the solution of (5.29)-(5.33); and go to (d),
d) Solve sub-problem (5.34)-(5.36) to determine the new realization of the uncertainties.
maxminT
yu
b y (5.34)
Subjected to: * ( 1) Fx Gy h Hu (5.35)
( 1), yy Ω u U (5.36)
e) Set * *T TUB c x b y where
*y is the solution of (5.34)-(5.36); and check if
UB LB
UB
. If the
86
condition is satisfied terminate the process, otherwise, add (5.37) and (5.38) to (5.29)-(5.33), then set
1 and go to (b).
( 1) ( 1)* Fx Gy h Hu (5.37)
( 1)Te b y (5.38)
Assuming that the worst realization of uncertainties occur at extreme points within the polyhedral
uncertainty set, the sub-problem (5.34)-(5.36) is re-formulated as a mixed-integer programming problem
shown in (5.39)-(5.52) by capturing the dual representation of inner minimization problem.
In order to determine the worst realization of the uncertain variables, binary variables ,db tu , ,
db tv , ,b tu ,
,b tv , ,l tu , ,g
b tu and
,g
b tv are introduced. Each pair of variables is mutually exclusive as shown in (5.47)-
(5.49). Moreover, an N-k contingency analysis is presented in (5.50), in which up to k components could be
unavailable in each period. The budget of uncertainty limits the combination of the binary variables that
yields the worst realization of the uncertain variables as shown in (5.51) and (5.52). Here, E and L are the
budget of uncertainty chosen by the decision maker.
4 ,0 4, , , , ,,,max 1 max 2
, , , ,,5 0 ,0 5 ,01 1 1, , , , , ,, ,
max 3,
ˆ
max
d d d db t b t b t b t b tNG NS NB b ts
i t s i i t i ti td di s b
b t b t b t b t b t b tb t b t
l l t
P u v P
P P I
P u v P
f
u,v,u ,vu ,u ,v
1
4 6 7, , , , ,
1
. 1
NT
tNL
l t l t l t l t l tl
u M u
(5.39)
1, 1i t
,: i tF (5.40)
1 1 2, , , , 0s
i i t i t s i tC NW ,: s
i tP (5.41)
2 2 3, , , ,
1
0NB
i t i t i b b ti
A
,: i tP
(5.42)
3 4, , 0b t b t
,: b tP (5.43)
, ,0 , , ,4 5, , , , , ,
( ).g m g m g m g m
b t b t b t b t b t b tNW C u v C
,,
:g m
b tP
(5.44)
87
4 6 7 3 4, , , , , ,
1
. 0NB
l b b t l t l t l t l tb
S
,: l tf (5.45)
6 7, , , ,
1
0NL
l b l t l l b l t ll
S X S X
: t
b (5.46)
, , 1d db t b tu v
(5.47)
, , 1b t b tu v (5.48)
, ,1
g gb t b t
v u (5.49)
,1
NL
l tl
u k
(5.50)
, , , , , ,1 1
NB NTg gd d
b t b t b t b t b t b tb t
u v u v v u E
(5.51)
,1 1
NL NT
l tl t
u L
(5.52)
The binary-to-continuous variable multiplication in the objective function (5.39) requires further
linearization as shown in (5.53)-(5.56) by introducing new auxiliary continuous variable . Here, is the
continuous variable and u is the binary variable of the binary-to-continuous multiplication, while M is a
large number which is considered as the upper bound of the continuous variable .
(1 )M u (5.53)
(1 )M u (5.54)
M u (5.55)
M u (5.56)
The solution to this problem is used to add more constraints (5.37), (5.38) to the master problem once the
condition at step (e) is not satisfied. Utilizing the solution of (5.39)-(5.52) and (5.57)-(5.59), a new
88
realization of uncertainties ( *u ) is captured by adding the constraints (5.37) and (5.38) to the master
problem.
,0, , , , ,,
ˆ ˆd d d d d db t b t b t b t b tb t
P P P u P v (5.57)
0, , , , , ,ˆ ˆb t b t b t b t b t b tu v
(5.58)
,0, , , , , ,
ˆ ˆg g g g g gb t b t b t b t b t b t
C C C u C v (5.59)
5.3 Case Study
In this section, two case studies were presented to show the effectiveness of the presented
approach as well as the impact of considerable penetration of microgrids on the maintenance scheduling
practices in the power system. The first case study uses a sample 6-bus power system while the second case
study utilize the modified IEEE-118 bus system.
5.3.1 6-Bus Power System
In this section, a 6-bus power system, which is composed of 3 thermal generation units and 7
transmission lines is utilized. The characteristics of the transmission line and the thermal generation units
are shown in Tables 5-1 and 5-2, respectively. The generation cost curve for the generation units is
piecewise linearized with four equal segments. The generation units G1, G2, and G3, are connected to
buses 1, 2, and 5 respectively. The minimum and maximum available periods as well as the required
maintenance periods for units G1-G3 were shown in Table 5-2. The maintenance cost for G1, G2 and G3
are $40,000, $30,000 and $10,000 respectively. The time step for this case study is one week and the
operation horizon is one year (52 weeks). The weekly system demand profile is shown in Figure 5-1.
In this case study, the contingency in transmission lines is ignored and the following cases are
considered:
Case 1 – Maintenance scheduling without aggregated microgrids
Case 2 – Maintenance scheduling with aggregated microgrids
Case 3 – Maintenance scheduling without aggregated microgrids considering the uncertainties
89
Case 4 – Maintenance scheduling with aggregated microgrids considering the uncertainties
Case 5 – Maintenance scheduling with aggregated microgrids knowing the probability distribution of
uncertainties
Table 5-1 Transmission line characteristics
Line
ID
From
Bus
To
Bus
Impedance
(p.u.)
Maximum
Power Flow
(MW)
L1 1 2 0.170 100
L2 1 4 0.258 100
L3 2 4 0.197 70
L4 5 6 0.140 60
L5 3 6 0.018 120
L6 2 3 0.037 150
L7 4 5 0.037 70
Table 5-2 Thermal units characteristics
Unit C1
($/MWh)
C2
($/MWh)
C3
($/MWh)
C4
($/MWh)
Pmax
(MW)
NHO
(Week)
MHO
(Week)
RHF
(Week)
G1 15 30 40 62.5 520 10 48 4
G2 20 32 46.25 73 360 10 48 5
G3 30 38 75 98 200 5 25 2
Figure 5-1 Demand profile for one year
5.3.1..1 Case 1 – Maintenance scheduling without aggregated microgrids
In this case, no microgrid is considered in the power system and the maintenance schedules of the
generation units are determined considering the nominal values for the demand in the power system. The
total maintenance and operation cost is $63.494M. In this case, unit G1 goes on maintenance in weeks 14-
17 when the demand is low; unit G2 goes on maintenance in weeks 42-46, and unit G3 goes on
maintenance in weeks 12-13 and 39-40. The total capacity on outage, in this case, is shown in Figure 5-2.
0
50
100
150
200
250
300
350
400
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51
Lo
ad (
MW
)
Period (Week)
90
Figure 5-2 Scheduled capacity on outage in Case 1
As shown in this figure, the minimum and maximum maintenance period for unit G3 are 5 and 25 weeks
respectively. Therefore, once G3 goes on maintenance in weeks 12 and 13, it needs to go on maintenance in
weeks 39 and 40.
In this case, the transmission lines L1, L2, L6 and L7 will be congested. By increasing the
capacity of these lines to twice of the capacity shown in Table 5-1, the maintenance scheduling pattern will
change as shown in Figure5-3. In this case, the total maintenance and operation cost will decrease to
$62.473M. As shown in this case, the congestion in the transmission line will impact the maintenance
schedule of the generation units as well as the total maintenance and operation cost.
Figure 5-3 Scheduled capacity on maintenance in Case 1 with the increase in the transmission line capacity
5.3.1..2 Maintenance scheduling with aggregated microgrids
Integrating microgrids in the power system provides higher flexibility at the demand side. Local
generation assets in microgrids can serve the demand once the locational marginal price (LMP) of
electricity falls beyond their marginal cost. In this case, the generation capacity of the microgrids is
91
assumed as 30% of the demand on each bus. The marginal cost of the local generation assets in microgrids
is 30 $/MWh. Here, the total operation and maintenance cost is $62.217M and units G1 and G2 are
scheduled for maintenance in weeks 14-17, and 40-44 respectively. Unit G3 goes on maintenance in weeks
9-10 and 34-35. The effect of microgrid on the net demand profile is shown in Figure 5-4. The net demand
is defined as the realized demand by the thermal generation units. In other words, the net demand is the
demand of the power system which is not served by the generation assets within the microgrids. As shown
in this figure, unit G1, which is the largest and cheapest available unit, will be on outage for maintenance in
periods that the net demand is low. Similarly, unit G2 goes on maintenance in the periods that the net
demand is reduced by utilizing the local generation of microgrids.
Figure 5-4 Demand profile with and without microgrids
The microgrids will change the net demand profile and affect the scheduled maintenance for unit
G3. As shown in this case, the scheduled maintenance is shifted from weeks 12-13 and 39-40 in Case 1, to
week 9-10 and 34-35 in this Case. Moreover, as a result of deviation in the net demand the maintenance
and operation cost is decreased. As the generation capacity in the microgrids increases, the maintenance
and operation cost will further decreases. Once the microgrids are completely self-sufficient (i.e. the total
demand can be served by local generation units in microgrids), the total maintenance and operation cost is
decreased to $62.155M and units G1 and G2 will be on planned outage for maintenance in weeks 13-16
and 21-25 respectively. Moreover, unit G3 is maintained in weeks 23-24 and 48-49.
5.3.1..3 Maintenance scheduling without aggregated microgrids considering the uncertainties
In this Case, the uncertainty in demand is considered as ±10% of the nominal values that are shown in
Figure 5-1, and the budget of uncertainty is 100 uncertain variables in the operation horizon. The total
0
50
100
150
200
250
300
350
400
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51
Lo
ad (
MW
)
Period (Week)
Demand (MW)
Net Demand with Microgrids
92
maintenance and operation cost, in this case, is $72.08M. The generation units G1 and G2 will be on outage
for maintenance in weeks 11-14, and 42-46 respectively. Unit G3 will be on outage for maintenance in
weeks 15-16, and 40-41.
By comparing Case 3 to Case 1, it is shown that the total operation and maintenance cost is
increased from $63.494M in Case 1 to $72.08M in Case 3 as a result of the introduced uncertainties in the
long-term operation horizon. Introducing the uncertainties will also affect the pattern of the generation
maintenance practices in the long-term operation horizon. As the budget of uncertainty decreases, the
operation cost will decrease because the decrease in the flexibility to determine the worst realization of the
uncertain variables. Once the budget of uncertainty is reduced to 30, the total maintenance and operation
cost is reduced to $66.736M. Figure 5-4 shows the total maintenance and operation cost with respect to the
determined budget of uncertainty. As shown in Figure 5-5, once the budget of uncertainty reached 240, the
total maintenance and operation cost is not changed. Moreover, the increase in the uncertainty interval will
lead to the increase in the total maintenance and operation cost. As shown in Figure 5-5, once the
uncertainty interval increases to ±15% of the nominal values, the total operation and maintenance cost will
increase. Similarly, reducing the uncertainty intervals will reduce the total operation and maintenance cost.
Figure 5-5 Increase in the total operation and maintenance cost because of the increase in the budget of
uncertainty
5.3.1..4 Maintenance scheduling with aggregated microgrids considering the uncertainties
In this Case, microgrids will serve the power system demand and the uncertainties associated with
the marginal cost of the local generation assets in microgrids as well as the uncertainties in the generation
capacity of microgrids are captured. The uncertainty bound for the marginal cost of generation assets in
microgrids is ±20% of the nominal marginal cost. The uncertainty bound for the local generation capacity
is ±10% of the nominal generation capacity within the microgrids. Once the budget of uncertainty is
60
65
70
75
80
85
0 20 40 60 80 100 120 140 160 180 200 220 240 260 280 300 320 340 360 380 400
Op
erat
ion a
nd
Mai
nte
nan
ce C
ost
($
)M
illi
ons
Budget of Uncertainty
Deviation=10%Deviation=15%Deviation=5%
93
considered as 100 uncertain variables, the total maintenance and operation cost of the system is $67.886M,
which is lower than that in Case 3. The scheduled outage for maintenance for units G1, G2 are in weeks 14-
17, and 40-44 respectively. Unit G3 will be on outage for maintenance in weeks 26-27 and 51-52. Similar
to Case 3, as the budget of uncertainty increases, the total maintenance and operation cost will increase. In
this case, as the budget of uncertainty increases to 420, the maintenance and operation cost increases to
$72.183M. Moreover, the maintenance schedule pattern for G1 and G2 are shifted to weeks 44-47 and 48-
52 and the maintenance of G3 is in weeks 15-16 and 42-43. By comparing Case 4 to Case 2, it is shown
that the total maintenance and operation cost increased from $62.217M to $67.886M as a result of
introducing the uncertainty sets in the risk-averse formulation. Furthermore, the generation maintenance
pattern was changed.
5.3.1..5 Maintenance scheduling with aggregated microgrids knowing the probability
distribution of uncertainties
In this case, the probability distribution functions of the uncertain variables are known and a risk-neutral
solution using stochastic programming is proposed for the generation maintenance scheduling problem.
The probability distribution function for electricity demand, the marginal cost of generation in the
aggregated microgrids, and the generation capacity of the aggregated microgrids are considered as normal
distribution functions with the mean values equal to the forecasted values and the standard deviation equal
to 3.33%, 6.66% and 3.33% of the mean values. In the proposed probability distribution functions, the
uncertain variables are between ±10%, ±20%, and ±10% of the mean values with 99.7% confidence
interval. In this case, the uncertainties are captured by generating 3,000 scenarios using Monte-Carlo
simulation and fast backward/forward method is used to reduce the number of effective scenarios to 12 by
eliminating the low probability scenarios and bundle the comparable scenarios [Dup03], [GAM].
The total maintenance and operation cost is $64.168M which is higher than that in Case 2
($62.217M) and lower than that in Case 4 ($67.886M). Moreover, in this case, the generation units G1 and
G2 were on maintenance in weeks 23-26, and 31-35 respectively. The generation unit G3 is on
maintenance in weeks 15-16, and 42-43.
94
5.3.2 IEEE 118-bus system
The modified IEEE 118-bus system with the total demand profile shown in Figure 5-6 is considered. The
system is composed of 54 generation units and 186 transmission lines. Generation units G4, G10, G11,
G27-G29, G36, G39, G40, G43-G45 with generation capacity over 300 MW are considered as candidates
for the maintenance. The marginal cost of the aggregated microgrids is shown in Figure 5-7. The capacity
of the local generation in microgrids is 30% of the demand on the network bus. Unlike previous case study,
the uncertainties in this case study include the contingencies in the transmission network. The candidate
lines for N-1 outage are L1, L10, L15, L38, L51, L90, L94, L103, and L126. The deviation of demand, the
marginal cost of the microgrid, and the local generation assets of the microgrids is ±10% of their respective
nominal values. Cases 1-4 that were introduced for the previous case study are considered here.
In Case 1, where there is no microgrid in the system and there is no contingency considered, the
total maintenance and operation cost is $649.422M and the planned outages for generation units were
shown in Table 5-3. Integrating microgrids in the power system in Case 2 decreases the total maintenance
and operation cost to $529.38M, which is lower than that for Case 1.
Figure 5-6 Demand profile for IEEE 118-bus power system
0
1000
2000
3000
4000
5000
6000
1 3 5 7 9 111315171921232527293133353739414345474951
Dem
and (
MW
)
Period (week)
95
Figure 5-7 Marginal cost of the local generation for the aggregated microgrids
The total maintenance and operation cost is further affected by the marginal cost (MC) of the generation
units in microgrids as shown in Figure 5-8. In this figure, the marginal cost of the microgrid aggregators is
changed by applying marginal cost multiplier (MCM). Here, with the increase in the marginal cost of the
local generation assets in microgrids, the generation units are used less frequent and the operation cost will
increase. Moreover, increasing the capacity of the local generation assets in microgrids can further affect
the net demand profile and reduces the total maintenance and operation cost of the power system. As
shown in Figure 5-8 with the increase in the capacity of the local generation assets with lower marginal
costs, the maintenance and operation cost decreases. As the marginal cost of the microgrids increases, the
effect of microgrids on the total maintenance and operation cost of the power system decreases. Similarly,
reduction in the installed capacity of low-cost local generation assets in microgrids will increase the total
operation and maintenance cost.
In Cases 1 and 2, no contingency in transmission line was considered and the demand is set to the
nominal values as shown in Figure 5-6. In Cases 3 and 4, the demand and the marginal cost of the
microgrids, as well as the capacity of the generation resources are considered as uncertain variables.
Moreover, the worst realization of the N-1 contingencies is also captured considering the budget of
uncertainty for the contingencies in the transmission network. The budget of uncertainty for the
contingency and other uncertainties in the operation horizon are 5 and 30 respectively.
0% 33% 66% 100%
Percentage of the full generation capacity
Marg
inal
Co
st (
$/M
Wh
)
15
30
45
96
Figure 5-8 Sensitivity of the total operation and maintenance cost to the installed generation capacity and
marginal cost multiplier (MCM) of microgrids
Table 5-3 Scheduled maintenance in weeks for cases 1-4
Component Case 1 Case 2 Case 3 Case 4
G4 14-18 14-18 13-17 42-46
G10 40-42 40-42 14-16 12-14
G11 44-46 11-13 40-42 14-16
G27 14-16 50-52 16-18 49-51
G28 41-44 17-20 43-46 41-44
G29 48-50 14-16 41-43 18-20
G36 50-52 42-44 50-52 48-50
G39 43-45 14-16 41-43 34-36
G40 17-20 39-42 14-17 2-5
G43 40-42 42-44 46-48 40-42
G44 13-16 45-48 44-47 45-48
G45 46-48 45-47 18-20 15-17
Table 5-4 The periods of transmission network contingencies for cases 1-4
Component Case 1 Case 2 Case 3 Case 4
L1 - - - -
L10 - - - -
L15 - - - -
L38 - - 2, 7, 8, 9, 10, 24, 25
29, 30, 31, 33 17
L51 - - 14, 46 12,30,14,16
L90 - - 5 21,22,26,12,2,9,10
L94 - - - 21, 19, 50, 43, 5, 14, 32,
48
L103 - - - 24,31
L126 - - - 29
0
100
200
300
400
500
600
700
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
Oper
atio
n a
nd M
ainte
nan
ce C
ost
($)
Mil
lions
Capacity of Local Generation in Microgrids (Percentage of Demand)
MCM = 0 MCM = 0.5MCM = 1 MCM = 1.5MCM = 2
97
Table 5-3 and 5-4 show the determined outages for maintenance and the outages considered to
yield the worst realization of the transmission network contingencies respectively. As shown in Table 5-3,
dispatching the local generation resources in microgrids will impact the maintenance schedule of the
generation assets in the power system. Moreover, microgrids can reduce the total maintenance and
operation cost of the power system by adjusting the net demand profile during the operation horizon. The
total maintenance and operation cost in Case 3 is $657.84M which is larger than that in Case 1 as a result of
uncertainties captured. The total maintenance and operation cost in Case 4 is $533.45M. Table 5-4 shows
the periods in which the contingencies in transmission network occurred for Cases 1-4. As shown in this
Table, most contingencies applied to lines L38, and L94 in Cases 3 and 4 respectively.
5.4 Summary
This chapter presents an approach for long-term maintenance scheduling in power systems
considering with large penetration of microgrids and the uncertainties in the operation horizon. The
uncertainty set captures the variation in the power system demand, the marginal cost of the microgrids, the
installed generation capacity within the microgrids, as well as the contingencies in the transmission
network. A two-stage robust optimization problem is formulated and column-and-constraint generation
procedure is proposed to solve the presented problem. A budget for uncertainty is considered to address a
trade-off between the conservativeness of the solution and the performance of the solution methodology.
The presented approach is applied to two case studies. The sensitivity of the total maintenance and
operation cost to the installed capacity of generation units in microgrids, the marginal cost of the
microgrids, and the considered budget of uncertainty was shown in a case study. It is shown that if the
marginal cost of the local generation in microgrids is small, leveraging the generation capacity of
microgrids to regulate the demand will decrease the total operation and maintenance cost of the power
system. Moreover, the total operation and maintenance cost is further decreased with the increase in the
capacity of the low-cost local generation assets in microgrids. It is also shown that the risk-averse solution
that captures the uncertainties in the long-term operation horizon will lead to higher maintenance and
operation cost compared to the deterministic solution. The presented risk-averse solution further compared
with risk-neutral solution knowing the probability distribution of the uncertain variables. It is shown that
98
the risk-averse solution will lead to higher operation and maintenance cost as the worst realization of the
uncertainties was captured.
99
CHAPTER 6
6COORDINATED OPERATION OF ELECTRICITY AND NATURAL GAS SYSTEMS: A CONVEX
RELAXATION APPROACH
In the electricity network, the unit commitment problem, which is solved by the independent system
operator (ISO), involves binary decision variables that represent the commitment of the generation units.
This problem has a discrete nonconvex feasibility region as a result of incorporated binary decision
variables. The contributions of this chapter are as follows: 1) develop an exact, tight, and computationally
inexpensive convex relaxation for the short-term operation problem in electricity and natural gas networks;
2) propose a consensus optimization framework for the coordinated operation of the electricity and natural
gas networks. The presented algorithm can be used by system operators to determine the tight convex
relaxation of the operation problems that lead to global solutions for the short-term operation of electricity
and natural gas networks. Furthermore, compared to earlier research [Sha05],[Gei07],[Mar12],[Cor14] that
assumed the electricity network operator has access to the natural gas network information, the presented
consensus optimization framework requires limited information being exchanged among the system
operators. The rest of the chapter is organized as follows, the problem formulation is presented in Section
6.1. The convex relaxation of the proposed problems and the solution methodology is presented in Section
6.2. A case study is presented in Section 6.3 to show the effectiveness of the proposed relaxations. The
summary is presented in Section 6.4.
6.1 Problem Formulation
In this section, the short-term operation problems for the natural gas and electricity networks are
formulated and the challenges for solving the coordinated operation of these systems using the consensus
optimization framework are discussed.
6.1.1 Short-term Operation of Natural Gas Network
The mathematical formulation for the natural gas network operation is presented in (6.1)-(6.8). The
100
objective is to minimize the natural gas consumption cost and the objective function is formulated as shown
in (6.1). The natural gas pipeline capacity constraint is given in (6.2). The natural gas supply volume at
each resource is further limited by (6.3). The operation limits on the natural gas pressure at each junction
are given in (6.4). The Weymouth equation presents the dependence of natural gas flow in the pipeline to
the pressure at the junctions on both sides of the pipeline as given in (6.5) and (6.6) [Men05],[Cfg]. The
nodal natural gas flow balance at each junction is illustrated in (6.7) where ( )H Pi i can be replaced by its
piecewise linear form. The natural gas compressor is modeled as (6.8) [Tom07].
,min .s GS j
s s tt
v
(6.1)
,max , ,max, , ,p p t pj o j o j of f f (6.2)
,s t ssv v v (6.3)
,j t (6.4)
( ).,
, , , ,, ,sgn ,p t p
j t o t j t o tj o j of C
(6.5)
1 , ,sgn ( )
1 , ,, ,,
j t o t
j t o tj t o t
(6.6)
, ,
, ,( ), ,, , ,s GS p P p P k HH i GGj f j t j j j
p t p t h H Ps t i i tj o j o k tsjA v f f P
(6.7)
, ,, ,, 1p p
j t o tj o j o
(6.8)
The presented problem is formulated as a mixed integer nonlinear optimization problem (MINLP). In
order to determine the decision on the natural gas flow direction in the pipeline, (6.5)-(6.6) are further
formulated as (6.9), where ,,t
j ou is a binary variable that is 1 if , ,j t o t and is 0 if , ,j t o t . Therefore,
the Weymouth equation represents a nonlinear non-convex constraint with discrete decision variables.
, ,, , , ,, , , ,2
p t p ptj t o t j t o tj o j o j o j of C u C
(6.9)
101
6.1.2 Short-term Operation of Electricity Network
The short-term operation problem for the electricity network is formulated in (6.10)-(6.17) as a unit
commitment problem. As shown in (6.10), the objective is to minimize the operation cost of the system.
The objective function is the summation of the operation costs of all generation units. The generation
dispatch of each generation unit is limited by the minimum and maximum capacity of the unit as shown in
(6.11). Minimum up/downtime constraints are given in (6.12)-(6.13). The ramp up/down limits are
enforced by (6.14)-(6.15). It is assumed that unit i sets at minimum generation iP prior to shutting down and
after starting up. The system-wide generation and demand balance is shown in (6.16). The power flow on
the transmission line l is limited by (6.17), considering the respective shift factors that relate the power flow
of the transmission line to the nodal injected power.
2, , ,
,min
i i
i i t i i t i i tP I t i
P P I (6.10)
Subjected to: , , ,. .i i t i t i i tP I P P I (6.11)
,( 1) ,( 1) , 0on oni t i i t i tX T I I
(6.12)
,( 1) ,,( 1)0
off offi t i tii t
X T I I
(6.13)
max, ,( 1) , ,( 1) , ,( 1)0 (1 ( )). ( )i t i t i t i t i i t i t iP P I I RU I I P (6.14)
max,( 1) , ,( 1) , ,( 1) ,0 (1 ( )). ( )i t i t i t i t i i t i t iP P I I RD I I P (6.15)
, ,D
b t i tb i
P P (6.16)
, ,. .
b
b b Dl l i t l b t l
b i B
F SF P SF P F
(6.17)
102
6.2 Proposed Convex Relaxation
6.2.1 Background
A theoretically strong convex relaxation that leverages the moment relaxation is proposed in [Las10].
The first order of moment relaxation represents the semi-definite relaxation of the original problem and as
the order of the moment relaxation increases, the asymptotic convergence to the global optimal solution for
the relaxed problem is guaranteed in polynomial time [La10]. However, the increase in the number of
variables in the moment relaxation matrices will lead to a considerable computational burden that impedes
the utilization of such relaxations in practice.
By employing the chordal property of the electricity network graph, a sparse moment relaxation is
formulated in [Mol15]. Despite the fact that the sparse formulation is a step forward to practically utilize
the theoretically perfect moment relaxation approach, the computational burden is still an obstacle even for
the sparse formulation presented for the medium-sized networks [Mol15]. The solution procured by
formulating lower order moment relaxations is computationally less expensive compared to those for
higher order moment relaxations; however, such solutions are not feasible for the original problem unless
the rank of the moment matrix is one. If the rank of a moment matrix is larger than 1, a higher order of
moment relaxation is required [Las10]. In such cases, the procured solution with a higher rank of the
moment matrix is a lower (upper) bound – yet infeasible solution – for the expected global optimal solution
of the minimization (maximization) problem. Since utilizing the higher order of moment relaxation is
computationally expensive, the alternative is to employ perturbation to procure a rank-1 solution for the
first order moment relaxation problem. However, the perturbation will lead to a rank-1 solution which
renders a feasible yet not necessarily optimal solution for the original problem. Another approach to
reducing the rank of the first order moment relaxation is to add valid constraints that include the elements
of the first order moment relaxation matrix [Koc16]. An example of such valid constraints is the
McCormick and disjunctive constraints. Although valid constraints will help to reduce the rank of the first
order moment relaxation matrix, the rank reduction may not lead to a rank-1 solution (i.e. the optimal
solution that is feasible for the original non-convex problem.) Therefore, it is vital to develop a
comprehensive approach to procuring a computationally inexpensive tight convex relaxation that renders a
103
rank-1 solution by incorporating the lessons learned from the earlier attempts.
As shown in the natural gas and electricity network operation problems, the binary decision variables
introduce non-convexity in the feasibility region. In order to handle these variables, the binary decision
variables are characterized by introducing constraints (6.18)-(6.19). For the sake of simplicity, the index of
hour t is eliminated from the presented formulation.
0 1rI (6.18)
2
r rI I (6.19)
By using (6.18)-(6.19), the constraints (6.9) and (6.11) with discrete decision variables are
transformed into nonlinear and nonconvex constraints with continuous decision variables ( rI ).
Incorporating (6.18)-(6.19) in (6.10)-(6.17) and (6.1)-(6.8) will lead to an NLP problem that is NP-hard
with no guarantee for global or local optimal solution in polynomial time. In the next section, tight convex
relaxations of the presented operation problems in electricity and natural gas networks will be formulated.
The solution to the formulated relaxed problems is the solution to the original problem.
6.2.2 Convex Relaxation of Natural Gas Network Operation Problem
The non-convexity associated with the operation of the natural gas network is because of introducing
constraint (6.9). Figure 6-1 shows the developed algorithm to solve the short-term operation problem for
the natural gas network. The steps of this algorithm are further described as follows:
Step (a) Define a reference vector of the first order monomials for the pressure of all junctions as given in
(6.20).
1 2[ .... ]NJ
π (6.20)
Step (b) Define monomials associated with the pipelines connected to each junction as given in (6.21).
Such monomials could be further expanded in the solution procedure.
j o j oj o j oj u u
, ,
( )[1 ]x y y y (6.21)
104
Figure 6-1 The algorithm to procure an optimal solution for the natural gas network operation problem.
Define a reference monomial vector associated with
natural gas pressures of all junctions (a)
Reformulate the natural gas transportation network
in SDP relaxed form (d), then solve the problem
Feasible
optimal
solution
Tightening the feasible region by adding RLT
constraints (e), then solve the problem
Tightening the feasible region by adding
McCormick and disjunctive constraints for element
of minors with higher ranks (g)
Add the elements of those 2by2 minors with higher
rank to the monomial vector associated with natural
gas pipelines connected to each junction (h)
Define the monomial vector associated with natural
gas pipelines connected to each junction (b)
Construct the sparse first order moment matrices
based on the monomial vector associated with
natural gas pipelines connected to each junction (c)
The ranks of all
moment matrices are 1?
The ranks of all 2by2
minors of the moment
relaxation matrix are 1? (f)
The ranks of all 2by2
minors of the moment
relaxation matrix are 1? (f)
No Yes
No
Yes
Yes
No
105
Here, (.)y represents the lifting variable corresponds to the nonlinear terms in the non-convex
optimization problem. For example, j o
y is the lifting variable that represents the nonlinear term
j o . Therefore, the nonlinear terms are substituted in the relaxed formulation using the corresponding
lifting variables (.)y . The general representation of the vector of monomials is given in (6.21). The first
element in this vector is 1 which represents the normalization of the monomial. The second vector is the
monomials associated with the direction of each pipeline that connects junctions j and o . It should be
noted that if the direction is a priori, the binary decision variable j o
u,
in monomials will be fixed. The third
vector of monomials represents the nonlinear relationship among nodal gas pressure at both sides of the
natural gas pipeline. This vector of monomials is formed for all pipelines that are connected to each
junction regardless of the decision on the direction of natural gas flow in them. The fourth monomial vector
is introduced to represent the binary to continuous terms associated with all the pipelines that are connected
to the junction j , where The elements of the monomial vector only capture the pipelines for which the
decision on the natural gas flow (j o
u,
) is a variable. Go to step (c).
Step (c) The moment matrix of the presented monomials for the junction j in (6.21), introduces a set of
lifting variables as presented in (6.22).
( ), ,
( () ), , , ,
) )( ( (), , ,
) )( ( ( () ), , , ,
1
2
u uj o j oj o j o
u u u uj o j o j o j oj o j o
u u uj o j oj o j o j oj o j o
u u u uj o j oj o j o j o j oj o j o
jj o
y y y
y y y yX
y y y y
y y y y
(6.22)
There are six types of independent variables in the moment matrix (6.22). Three of them (i.e. j o
u
,
y
,j o
y , and u j o j o
( ),y ) are the lifting variables given in (6.21). Two of them (i.e.
j and
o ) are given in
step (a) that are utilized in various moment matrices of each junction as represented by (6.20). The last
variable (i.e. u j oj o )( ,
y ) is a lifting variable that is introduced in (6.22) to complete the moment matrix.
Go to step (d).
106
Step (d) By utilizing the lifting variable that was introduced in step (c), the SDP relaxation for the short-
term operation of the natural gas network given in (6.1)-(6.8) is reformulated as (6.23)-(6.28). Here the
decision variables are the elements of the moment matrix (6.22) as well as the vector of pressure given in
(6.20).
,min . ( )s GS j j o
pss j j o
o
A C
y (6.23)
,
,max ,max, , ,( )
( 2 )j oj o j o
p p pj o j o j ou
f C f
y y (6.24)
, ( )j o
psj sj os
o
v A C v
y
(6.25)
, ,, 1p p
j j oj o j o (6.26)
,
,
( ) ( 2 )( ), , ,
( 2 ) ( )( ), ,
s GS p Pj f j
p P k HH i GGt j j j
p psA C C uj j o j oj o j o j o j oo
p hC H Pu i ij o j o j o j o kP
y y y
y y
(6.27)
0jX (6.28)
The presented SDP problem is solved to determine the optimal solution for the original problem. If the
rank of all sparse moment matrices associated with each junction is one, the procured optimal solution is
feasible for the original non-convex problem (6.1)-(6.8). Otherwise, the algorithm proceeds to step (e) to
further tighten the presented relaxation and to render rank-1 moment matrices.
Step (e) In this step, a tighter SDP relaxation is formulated by leveraging reformulation-linearization
technique (RLT). The convex-relaxed problem (6.23)-(6.28) is presented in a compact form as shown in
(6.29)-(6.34). The nodal balance constraint given in (6.27) is converted into two inequalities as shown in
(6.30), (6.31). The inequalities given in (6.24)-(6.26) are also presented in the general form in (6.30)-(6.31).
RLT is used to form valid constraints (6.32) and (6.33) to tighten the relaxation, while the moment matrices
defined in step (c) (i.e. j
X ) and the monomials are given in step (b) (i.e. j
x ) are utilized to formulate these
107
constraints.
,
minj j
Tj j
j
x Xξ x (6.29)
1 1 0j j
j g x h (6.30)
2 2 0j j
j g x h (6.31)
1 1 1 11 1 1 10
T T T Tj j j jTj j j jj j j g X h x g x hg g h h (6.32)
2 2 2 22 2 2 20
T T T Tj j j jTj j j jj j j g X h x g x hg g h h (6.33)
0jX (6.34)
Step (f) Solving the SDP relaxation problem with additional valid constraints will reduce the rank of the
moment matrices corresponding to the procured solution (i.e. the lower bound to the optimal solution.)
However, the rank of some of the sparse moment matrices may remain higher than one. To add a new set of
valid constraints, it is necessary to exploit those elements of the moment matrix that result in the solution
with a rank higher than 1. Theorem 1 is employed to identify such elements in moment matrices.
Theorem 1. The rank of a moment relaxation matrix is one, if and only if the rank of all 2by2 minors (e.g.,
( , ),( , )i k n mjX where i, k, n, m are the indices for the rows and columns of jX ) of that matrix are one,
( , ),( , )( ) 1 ( ) 1 , , ,
i k n mj jrank rank i k n m X X . The proof of this theorem is discussed in [Koc16].
In this step, the ranks of all 2by2 minors are checked. If the ranks of all 2by2 minors are one, the convex
relaxation is tight and the optimal solution for the original problem is procured. Otherwise, if the previous
step was step (e) then go to step (g); if the previous step was step (g) then go to step (h).
Step (g) Once the minors with a rank higher than 1 are identified in step (f), McCormick constraints and
disjunctive cuts are employed to reduce the rank of each minor within any moment matrix. The McCormick
constraints for a bi-linear term xy are given in (6.35)-(6.38), where the upper and lower bounds of variable
x and y are given as x y, and x y, respectively. This technique is applied to each term associated with the
higher than rank-1 minors of each moment matrix.
xy x y x y x y (6.35)
108
xy x y x y x y . (6.36)
xy x y x y x y (6.37)
xy x y x y x y (6.38)
A list of such valid constraints are given in (6.40)-(6.43) and in the Appendix, where the scalar
lifting variables that are used as the elements of the moment matrix (6.22) are captured. As an example
here, the upper and lower bounds of ( ),u j o j o
y
in the 2by2 minor given in (6.39) are presented in (6.40)-
(6.43).
)( ,
( () ), ,
1u j o
u uj o j oj o j o
y
y y
(6.39)
Here, the lower and upper bounds of the off-diagonal elements in the 2by2 minor are ( ),
0 1u j o
y and
j o
y
( )
0 . Using (6.35) and considering the lower bounds as zero for the off-diagonal elements,
(6.40) is formed. All elements of the moment matrix are nonnegative. Similar to (6.36), the upper bounds
of the off-diagonal elements of the 2by2 minor (6.39) are used in (6.41) to form a lower bound for the
diagonal element (( ),u j o j o
y
) in the 2by2 minor shown in (6.39). The upper bounds of the element
( ),u j o j oy
are enforced in (6.42)-(6.43) using the general form (6.37)-(6.38). As
u j oy
( ),0 1 , the upper
bound of( ),u j o j o
y
is j o
y ( )
as given in (6.42). Using (6.37), j o
y ( )
and its upper bound ( ) are
multiplied by zero which is the lower bound of u j o
y )( ,. Similarly,
j oy
( ) is multiplied by 1 which is the
upper bound of u j o
y )( ,. Using (6.38),
u j oy )( ,
and its upper bound (i.e. 1) are multiplied by zero which is the
lower bound ofj o
y ( )
. Therefore, the only non-zero term in (6.43) is u j o
y )( , multiplied by which is
the upper bound ofj o
y ( )
. The rest of the valid constraints are presented in the Appendix.
u j o j oy
( ),0 (6.40)
109
u uj o j oj o j o
y y y
( ), ,
( )(1 ) (6.41)
( ),u j o j o j oy y
(6.42)
u uj o j oj o
y y
( ), ,
( ) (6.43)
In this step, only the valid constraints associated with the 2by2 minors with a rank higher than 1 – that
were determined in step (f) – will add more cutting planes to the feasible region of the problem to further
tighten the relaxation. Go to step (f) to further check the rank of the 2by2 minors.
Step (h) The elements of the 2by2 minors with a rank higher than 1 will be added as new
monomials to the vector of monomials in step (b). Traditionally, it is necessary to employ a higher order
moment matrix to procure a rank-1 solution [Las10]. However, in this step, by applying Theorem 1, only
certain elements that are present in the 2by2 minors with a rank higher than 1, will be added to the vector of
monomials to tighten the feasible region of the procured moment relaxation. Therefore, instead of adding
all elements of the higher order moment matrix for each junction, adding a limited number of monomial
will not increase the size of the moment matrix dramatically. Go to step (b).
6.2.3 Convex Relaxation of the Electricity Network Operation Problem
The algorithm developed for the convex relaxation of this problem is similar to the algorithm
developed for the natural gas network operation problem. The objective is to determine a tight convex
relaxation for the electricity network operation problem considering the binary decision variables that
represent the commitment of the generation units. The following steps are taken to achieve this objective.
Step (a) A vector of monomials is introduced as shown in (6.44).
NG NG
I I P Py y y y
1 1
[1 ... ... ]z (6.44)
Step (b) The associated first-order moment matrix of the presented monomials is given in (6.45) in which
the binary-to-continuous terms are incorporated.
110
1 1
1 1 1 1 1 1
1 1
21 1 1 1 11
21 1
1 ... ...
... ...
... ... ... ... ... ... ...
... ...
... ...
... ... ... ... ... ... ...
... ...
NG NG
NG NG
NG NG NG NG NG NG
NG NG
NG NG NG NG NG NG
I I P P
I I I I I P I I
I I I I I P I P
P I P I P P PP
P I P I P P P P
y y y y
y y y y y
y y y y y
y y y y y
y y y y y
Z
(6.45)
Step (c) The lifting variables introduced in (6.45) are utilized to reformulate the electricity network
operation problem (6.10)-(6.17) to (6.46)-(6.51). Here, the inequality constraints presented in (6.48)-(6.49)
represent nodal power balance in the electricity network. If the rank of the moment matrix (6.45) is one, the
procured lower bound solution is an optimal solution for the original problem and the algorithm will
terminate, otherwise, proceed to step (d).
21
minii
i i P i IPZ i
y y y (6.46)
Subjected to: . .i i ii I P i IP y y P y (6.47)
i i
Db I P
b i
P y (6.48)
'
2
'i i
Db PP
b i i
P y
(6.49)
. .i
b
b b Dl l P l b l
b i B
F SF y SF P F
(6.50)
0Z (6.51)
Step (d) Add the RLT constraints to tighten the procured relaxation. If the tightened relaxed problem
renders a rank-1 solution, the procured lower bound is the optimal solution for the original problem.
Otherwise, employ Theorem 1 to identify the minors with a rank higher than 1 and proceed to step (e).
Step (e) Once the 2by2 minors with a rank higher than 1, were identified, employ McCormick and
disjunctive constraints to reduce their rank. This would also reduce the rank of the moment matrix
introduced in step (b). By using these valid constraints, the size of the problem will not increase, however,
the procured solution would be optimal for the original problem. If the relaxed problem that is tightened by
111
adding the valid constraints procures a rank-1 solution, the algorithm is terminated. Otherwise, identify the
2by2 minors with higher than rank-1 and proceed to step (f).
Step (f) The elements of the minors with a rank higher than 1, are added to the vector of monomials
introduced in step (a).
Using the proposed approach, the feasibility of the optimal solution to the convex-relaxed electricity
network operation problem is guaranteed. With the increase in the order of the moment relaxation, the
solution for the convex-relaxed problem will converge to the optimal solution of the original problem
[Men05]. The proposed algorithm is computationally less expensive compared to employing higher order
moment relaxation as it will only add the necessary elements to the vector of monomials to construct the
moment matrix.
6.2.4 Coordination among Electricity and Natural Gas Networks Operation
Here, a consensus optimization framework is developed to capture the interactions among the natural gas
and electricity networks. The presented consensus optimization is solved using the alternating direction
method of multipliers (ADMM) [Boy11]. The vector of shared variables among the natural gas and
electricity networks is the volume of the natural gas demand for the natural gas generation units. The
outline of the proposed problem is shown in (6.52)-(6.54). Here, ( )f x represents the objective function of
the natural gas operation problem given in (6.23) and ( )g z represents the objective function of the electricity
network operation problem given in (6.46). The convex-relaxed problems procured in sections III.B and
III.C are represented by (6.52) and (6.53) respectively. The feasible regions are shown as x and z
respectively, and λ is the Lagrange multiplier of the consensus constraint.
( ) . . , :xMin f x s t x x z (6.52)
( ) . . , :zMin g z s t z z x (6.53)
The consensuses equality constraint is relaxed with respective Lagrange multiplier in the
augmented Lagrangian function shown in (6.54) and is a scalar.
112
2
2( , , ) ( ) ( ) ,
2L x z f x g z x z x z
(6.54)
In order to solve this problem, (6.52)-(6.53) are decomposed into two separate problems. The first
problem (6.55) is the x-update problem and the second problem (6.56) is the z-update problem. As shown
in (6.55), the x-update problem is solved in each iteration, while the variables associated with the z-update
problem are fixed. Similarly, in the z-update problem (6.56) is solved while the variables associated with
the x-update problem are fixed. Here, the x-update problem is the convex-relaxed natural gas network
operation problem and the z-update is the convex-relaxed electricity network operation problem. The
vector of Lagrange multipliers for the consensus constraints is updated in each iteration as shown in (6.57).
1 arg min ( , , )xx
x L x z
(6.55)
1 1arg min ( , , )zz
z L x z
(6.56)
1 1 1( )x z (6.57)
6.3 Case Study
A sample 6-bus electricity network and 7-junction natural gas network are considered as shown in Figure
6-2. The 6-bus electricity network is composed of 2 natural gas generation units (G1 and G2) and 1 fossil
fuel generation unit (T1). The characteristics of the generation units and transmission lines are shown in
Tables 6-1 and 6-2, respectively. The fuel price of the thermal unit is 3.5 $/MMBtu. The characteristics of
the Natural gas resources and pipelines are shown in Tables 6-3 and 6-4, respectively. The minimum and
maximum pressures at junctions of the natural gas network are 105 and 170 Psig respectively. The natural
gas demands at junctions J1, J2, and J3 are 15%, 50% and 35% of the total natural gas demand,
respectively. The electricity demand on buses 3, 4, and 5 are 20%, 40%, and 40% of the total electricity
demand, respectively. The normalized demand profiles for the electricity and natural gas are given in
Figure 6-3. The peak demands for electricity and natural gas network are 256 MW at hour 17 and 6780 kcf
113
at hour 20, respectively. Here, the number of binary variables in the natural gas and electricity network
operation problems is 24 and 72 respectively. The number of continuous variables for the natural gas and
electricity network operation problems is 504 and 216 respectively. The following four scenarios are
presented:
Scenario 1) Uncoordinated operation of the electricity and natural gas networks
Scenario 2) Coordinated operation of the electricity and natural gas networks
Scenario 3) Scenario 2 with congestion in the electricity network
Scenario 4) Scenario 3 with congestion in the natural gas
Figure 6-2 The 6 bus-system interconnected with 7-junction natural gas
Figure 6-3 The normalized demand profile for electricity and natural gas
PL6J4
C1
PL2J2
J6PL5
J5
PL4J3
PL3
PL1J7
123
456
L1
L2
L3
L4L5
L6L7
G1
G2
T1
v1
v2
GT Gas resourceCompressorJunctionGas-fired unitsThermal units
J1
0.5
0.6
0.7
0.8
0.9
1
1.1
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
No
rmal
ized
Dem
and
Pro
file
Time (Hours)
Electricity Demand Profile Natual Gas Demand Profile
114
Table 6-1 Generation unit charcteristics
Unit
a
(MMBtu/MW
h2)
b
(MMBtu/MW
h)
c
(MMBtu/h
)
Pmin
(MW)
Pmax
(MW)
RU/RD
(MW/h)
,offon
i iT T
(h)
G1 0.028 12.196 170.48 10 320 100,100 2,2
T1 0.0676 18.543 289.9 50 160 50,50 4,3
G2 0.058 10.69 136.29 10 60 50,50 1,1
Table 6-2 Transmission line characteristics
Line
ID
From
Bus
To
Bus
Impedance
(p.u.)
Ling Rating
(MW)
L1 1 2 0.170 80
L2 1 4 0.258 80
L3 2 4 0.197 70
L4 5 6 0.140 60
L5 3 6 0.018 80
L6 2 3 0.037 150
L7 4 5 0.037 70
Table 6-3 Natural gas resources characteristics
Resource Minimum Capacity
(kcf/h)
Maximum Capacity
(kcf/h)
Cost
($/kcf)
v1 1500 5000 2.6
v2 1000 6000 3.2
Table 6-4 Natural gas pipeline characteristics
Line
ID
From
Junction
To
Junction
Pipeline
Constant (kcf
/Psig)
Compressor
factor
PL1 J1 J2 51 0
PL2 J2 J4 0 1.21
PL3 J2 J5 83 0
PL4 J3 J5 64 0
PL5 J5 J6 44 0
PL6 J4 J7 63 0
6.3.1 Scenario 1 – Uncoordinated operation of the electricity and natural gas networks
The hourly unit commitment solution is presented in Table 6-5. At peak demand in hour 17, the
generation dispatches of G1 and G2 are 196 MW and 60 MW, respectively. The total operation cost of the
electricity network is $298,722. The volumes of natural gas withdrawn from resources v1 and v2 at hour 17
are 5000 kcf and 5827.7 kcf respectively. The total operation cost of the natural gas network is $675,887
and junction pressures of this network are shown in Table 6-6.
115
Table 6-5 Hourly unit commitment in Scenario 1
Unit Hours (0-24)
G1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
T1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
G2 1 1 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
Table 6-6 Natural gas pressures in scenario 1 at hour 17
Junction ID J1 J2 J3 J4 J5 J6 J7
Pressure
(Psig) 131.44 137.80 105.00 137.80 137.55 178.42 165.96
The increase in the natural gas supply to serve the natural gas generation units leads to the violation of the
operation limits of the natural gas network. For example, to serve the 3565.58 kcf natural gas demand of
G1, the pressure at junction J6 is increased to 178.42 psig, which is higher than the maximum limit. Thus,
the natural gas network is unable to supply the natural gas generation units.
6.3.2 Scenario 2 – Coordinated operation of the electricity and natural gas networks
In this scenario, the operational planning of the electricity network is coordinated with that for the
natural gas network. However, the capacity limits of the transmission lines and pipelines are relaxed to
avoid any congestion. In this scenario, in addition to units G1 and G2, T1 which is a more expensive unit is
also committed as shown in Table 6-7. Compared to scenario 1, unit T1 is committed at hours 16-20, and
G2 is committed at hour 7.
At hour 17, all units are committed to serve the peak electricity demand. The generation dispatch of G1,
T1 and G2 is 153.5 MW, 50 MW, and 52.5 MW respectively. The generation dispatches of natural gas
units G1 and G2 are decreased compared to those in scenario 1. Furthermore, the volumes of natural gas
withdrawn from resource v1 and v2 are 4883.82 kcf and 4871.62 kcf respectively. The junction pressures in
the natural gas network are given in Table 6-8. In this scenario, the shared variables among the electricity
and natural gas networks are the consumption volumes of the natural gas generation units. At hour 17, the
natural gas consumption volumes for G1 and G2 are 2701.97 MMBtu and 1067.38 MMBtu, respectively.
The flow of natural gas in pipeline PL3 is 113.83 kcf/h from junction J5 to J2 and the junction pressures are
within their operation limits.
116
The total operation cost of the electricity network is increased to $303,956 which is 1.75% higher
than that in scenario 1. Furthermore, the total operation cost of the natural gas network is decreased to
$661,132.
Table 6-7 Hourly unit commitment in Scenario 2
Unit Hours (0-24)
G1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
T1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 0 0 0 0
G2 1 1 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
Table 6-8 Natural gas pressures in scenario 2 at hour 17
Junction ID J1 J2 J3 J4 J5 J6 J7
Pressure
(Psig) 122.91 128.76 105.00 128.76 128.76 170.00 150.19
In order to show the performance of the presented algorithm the characteristics of the sparse moment
matrices associated with each junction of the natural gas network were considered. The procured convex
relaxation is exact if the rank of the all moment relaxation matrices is one. If the rank of a matrix is one, all
of its eigenvalues are zero but one. Alternatively, if the ratio of the largest eigenvalue to the second largest
eigenvalue of the moment matrix is very large (close to infinity), the rank of the matrix is one. Therefore,
this ratio is used as a measure of the tightness for the proposed relaxation. The ratio of the two largest
eigenvalues of the moment matrix associated with each junction of the natural gas network is evaluated for
all hours in the operation horizon. As an example, this ratio is given for scenario 2 at peak hour (hour 17) in
Table 6-9. Here, the rank-1 moment relaxation matrix corresponds to junction J4 is procure in step (d) of
the algorithm before adding any RLT constraints, or new variables to the vector of monomials. However,
the rank-1 moment relaxation matrices correspond to junctions J6 and J7 are procured after adding the RLT
constraints in step (e). Once the valid constraints added in step (g), the rank-1 moment relaxation matrix is
procured for junctions J1 and J3. Finally, the rank-1 moment relaxation matrix is procured at step (h) for
junctions J2 and J5 for which the variables associated with the rank higher than 1 2by2 minors are added to
the vector of monomials in addition to the constraints added in the previous steps. YALMIP [Lof04] is used
to solve the problem and Mosek 7 is used as the SDP solver on a PC with 3.2 GHz Intel i5 processor and 8
GB of memory. The computation time at the last iteration of the ADMM approach to reach the optimal
117
solution for 7-junction natural gas and 6-bus electricity network operation problems is 14.65 sec and 100.62
sec respectively. The total computation time to converge to a solution using the ADMM approach is 829.94
sec.
Table 6-9 The ratio of the two largest eigenvalues in scenario 2 at hour 17
Junction
ID J1 J2 J3 J4 J5 J6 J7
Step (d) 5.7E1 2.1E0 4.3E1 1.7E9 1.8E0 6.4E2 3.8E2
Step (e) 7.8E2 1.6E1 1.5E2 - 2.3E1 9.8E9 1.5E10
Step (g) 7.9E12 7.3E3 9.2E10 - 6.5E2 - -
Step (h) - 9.4E11 - - 8.3E9 - -
6.3.3 Scenario 3 – Scenario 2 with Congestion in the Electricity Network
In this scenario, the capacity limits of the pipelines are relaxed to avoid congestions in the natural
gas network, and the power flow limits given in Table 6-2 are enforced for the electricity network. As
shown in Table 6-10, the generation units are committed for more number of hours compared to scenario 2
to avoid congestion in the electricity network. Compared to scenario 2, the more expensive generation unit
T1 is committed at hours 10-15 and 21-22 while the natural gas unit G2 is committed at hours 2-6. The
total operation cost of electricity network is increased to $313,817, which is 3.24% higher than that in
scenario 2.
Table 6-10 Natural gas pressures in scenario 3 at hour 17
Unit Hours (0-24)
G1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
T1 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 0 0
G2 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
The decrease in the total dispatch of the natural gas generation units leads to the decrease in
consumption in the natural gas network. The total operation cost of the natural gas network is decreased to
$631,587 which is 4.46% less than that in scenario 2. At this hour, the volumes of natural gas withdrawn
from the resources v1 and v2 are 5000 kcf and 4587.1 kcf, respectively. The total volume of natural gas
withdrawn is decreased compared to scenario 2 as the overall natural gas consumption of the natural gas
generation units is decreased. The junction pressures in the natural gas network at hour 17 are given in
Table 6-11. The natural gas consumptions for G1 and G2 at hour 17 are 1792.3 MMBtu and 272.81
118
MMBtu, respectively. Compared to scenario 2, it is shown that congestion in the electricity network leads
to the changes in the demand of the natural gas generation units. At hour 17, the flow of natural gas in
pipeline PL3 is 554.15 kcf/h from junction J5 to J2 and the junction pressures are within the operation
limits.
Table 6-11 Natural gas pressures in scenario 3 at hour 17
Junction ID J1 J2 J3 J4 J5 J6 J7
Pressure
(Psig) 118.71 125.72 105.0 125.72 125.9 169.6 145.28
The ratio of the two largest eigenvalues of the moment relaxation matrix associated with each junction of
the natural gas network at hour 17 in scenario 3 is considered as a tightness measure for the proposed
convex relaxation as given in Table 6-12. Here, the rank-1 moment relaxation matrix corresponds to
junction J4 is procured in step (d) of the algorithm. Adding RLT constraints in step (e) would lead to the
rank-1 moment relaxation matrices correspond to junctions J6 and J7. The rank-1 moment relaxation
matrices correspond to junctions J1, J2 and J3 are procured after adding the related valid constraints in step
(g). Finally, the rank-1 moment relaxation is procured in step (h) for junction J5, where the variables
associated with higher than 1 rank 2by2 minors are added to the vector of monomials in addition to the
constraints added in previous steps. Compared this scenario with scenario 2, it is shown that the rank-1
moment relaxation matrix corresponding to junction J2 is procured in step (g) in scenario 3 while that was
procured in step (h) in scenario 2.
Table 6-12 The ratio of the two largest eigenvalues in scenario 3 at hour 17
Junction ID J1 J2 J3 J4 J5 J6 J7
Step (d) 9.2E2 2.1E1 6.5E1 9.8E9 2.1E1 4.1E3 2.5E3
Step (e) 1.3E3 2.2E2 5.1E3 - 9.7E1 6.9E8 2.6E9
Step (g) 1.1E14 9.4E9 8.3E11 - 4.7E4 - -
Step (h) - - - - 7.4E8 - -
6.3.4 Scenario 4 – Scenario 3 with Congestion in the Natural Gas Network
In this scenario, the capacity of pipeline PL3 is limited to 500 kcf/h while the capacity limits of the
transmission lines are also considered. The congestion in the natural gas network will change the
generation dispatch for some hours but the commitment status of the generation units remains the same as
119
scenario 4. The congestion in the natural gas network affects the volume of natural gas withdrawn from the
resources. As the natural gas demand of G1 alleviates the congestion in the natural gas network, the natural
gas price for G1 is lower than that for G2 in this scenario. Therefore, the total operation cost of electricity
network is decreased to $278,239 which is 11.33% less than that in scenario 3. While the generation
dispatch and natural gas demand of natural gas generation units remain the same as those in scenario 3, the
total operation cost of the natural gas network is increased to $634,025 which is 0.39% more than that in
scenario 3. At hour 17, the volume of the natural gas withdrawn from resource v1 is decreased to 4945.85
kcf and the volume of the natural gas withdrawn from resource v2 is increased to 4641.24 kcf. Although
the resource v1 is cheaper compared to v2, the limited capacity of pipeline PL3 (500 kcf/h) will reduce the
withdrawn volume of natural gas from this resource. The pressures at the junctions at hour 17 are given in
Table 6-13. The efficiency of the presented algorithm for tightening the convex relaxation in this scenario
is shown in Table 6-14.
Table 6-13 Natural gas pressures in scenario 4 at hour 17
Junction ID J1 J2 J3 J4 J5 J6 J7
Pressure
(Psig) 118.75 125.75 105.0 125.75 125.90 168.78 145.74
Table 6-14 The ratio of the two largest eigenvalues in scenario 4 at hour 17
Junction
ID J1 J2 J3 J4 J5 J6 J7
Step (d) 7.6E2 3.4E1 1.5E0 8.2E9 9.4E1 3.9E3 3.5E3
Step (e) 9.7E2 3.8E2 6.1E3 - 2.2E2 7.4E8 7.2E9
Step (g) 2.7E13 2.3E10 9.8E10 - 6.9E8 - -
Step (h) - - - - - - -
6.3.5 Discussion
The generation profile of the generation units G1, G2, and T1 in scenarios is presented in Figure 6-4,
Figure 6-5, and Figure 6-6, respectively. By comparing scenarios 1 and 2, it is shown that the coordination
among the electricity and natural gas networks will decrease the generation dispatch of the natural gas
generation units at hours where the natural gas demand is increased in the network. During hours 16-20 the
generation dispatch of G1 is decreased in scenario 2 compared to scenario 1 as shown in Figure 6-4; and
the generation dispatch of T1 is increased as shown in Figure 6-6. A comparison of scenarios 2 and 3
120
highlights the impact of congestion within the electricity network on the dispatch profile of the generation
units. As shown in Figure 6-4, the generation dispatch of G1 is decreased during hours 2-6 while the
generation dispatch of G2 is increased in the same period as shown in Figure 6-5. Additionally, the
generation dispatch of T1 is increased during hours 10-18. Comparing scenarios 3 and 4 shows the impact
of congestion in the natural gas network on the generation dispatch of the generation units in the electricity
network. As congestion in the natural gas network will decrease the marginal cost of the generation unit G1
during hours 2-6, the generation dispatch of G1 will increase compared to that in scenario 3. The generation
dispatch of G2 decreases accordingly. By comparing the generation dispatch of G1 for scenarios 2, 3, and
4, it is shown that although congestion in the natural gas network will increase the dispatch of G1, the
limited capacity of the transmission lines in the electricity network will limit the increase in the generation
dispatch to avoid congestion in the electricity network.
Furthermore, the presented formulation captures the variability in the electricity and natural gas network
by addressing the bi-directional flow of natural gas in the pipelines. Here, the flows of natural gas in
pipeline PL3 in scenario 3 and scenario 4 are given Figure 6-7. As the flow of natural gas in this pipeline is
limited to 500 kcf/h in scenario 4, the variation in the generation dispatch of the natural gas generation units
in Figure 6-4, 6-5, and 6-6 will result in multiple changes in the direction of the natural gas flow in the
pipeline PL3 in the operation horizon. The presented framework procures the flow direction in the pipelines
as a result of the variation in the generation dispatch of the natural gas generation units.
Moreover, the congestion in the natural gas and electricity networks will impact the volume of
natural gas withdrawn from the natural gas resources. The hourly volumes of natural gas withdrawn from
the sources v1 and v2 in scenarios 3 and 4 are shown in Figure 6-8. In congestion period for natural gas
network in scenario 4, the natural gas withdrawn from source v1 is decreased compared to that in scenario 3
and the natural gas withdrawn from source v2 increased to compensate for the shortage in the natural gas
supply. As a result of congestion in the natural gas network, the total operation cost of this network is
increased from $631,587 in scenario 3 to $634,025 in scenario 4. As the increase in the natural gas demand
of G1 mitigates the congestion in the natural gas network, the dispatch of this generation unit is increased
as shown in Figure 6-4.
121
Figure 6-4 The generation dispatch of G1 in scenarios
Figure 6-5 The generation dispatch of G2 in scenarios
Figure 6-6 The generation dispatch of T1 in scenarios
100
120
140
160
180
200
220
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Gen
erat
ion D
isp
atch
[M
W]
Time [Hours]
Scenario 1 Scenario 2 Scenario 3 Scenario 4
0
10
20
30
40
50
60
70
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Gen
erat
ion D
isp
atch
[M
W]
Time [Hours]
Scenario 1 Scenario 2 Scenario 3 Scenario 4
0
10
20
30
40
50
60
70
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Gen
erat
ion D
isp
atch
[M
W]
Time [Hours]
Scenario 1 Scenario 2 Scenario 3 Scenario 4
122
Figure 6-7 The flow of natural gas in pipeline PL3 in scenarios 3-4.
Figure 6-8 The profile for the natural gas volume withdrawn from sources in scenarios 3-4.
6.3.6 IEEE 118-Bus System with 12-junction Natural Gas network
In this case, a modified IEEE-118 bus system is supplied by a 12-junction natural gas network. The
electricity network has 46 fossil fuel generation units, 8 natural gas generation units, 186 branches, and 91
demand entities. The total capacity of the natural gas generation is 575 MW. The peak load is 3700 MW
which occurs at hour 21. The natural gas network that supplies the electricity network is composed of 12
junctions, 12 pipelines, two compressors, and 12 natural gas demand entities including the natural gas
generation units as shown in Figure 6-9. The natural gas peak demand of 14,500 kcf that occurs at hour 19.
Here, the number of binary variables for the natural gas and electricity network operation problem is 168
-1500
-1000
-500
0
500
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Nat
ura
l G
as F
low
[kcf
/h]
Time [Hour]
Scenario 3 Scenario 4
2200
2700
3200
3700
4200
4700
5200
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Nat
ura
l G
as F
low
[kcf
/h]
Time [Hour]
v1 in Scenario 4
v2 in Scenario 4
v1 in Scenario 3
v2 in Scenario 3
123
and 3888 respectively. The number of continuous variables for the natural gas network operation problem
is 648 and the number of continuous variables for the electricity network operation problem is 3888. In the
last iteration of the ADMM approach, the computation time for the 12-junction natural gas network
operation problem is 20.94 sec. Here, the proposed algorithm for the natural gas operation problem is used
to determine the sparse formulation for the IEEE 118-bus network operation problem. The computation
time for the large-scale electricity network is reduced by leveraging the sparse formulation of the unit
commitment problem and the solution time is 160.1 sec.
Figure 6-9 12-junction natural gas network
In this case study, the uncoordinated operation scheme (scenario 1) is compared to the
coordinated operation scheme (scenario 2). As the natural gas network is constrained by the operation
limits on the junction pressures, the natural gas generation units may not operate at their full capacity. As a
result of imposing the pressure limits at junction J3, the dispatches of units G4 and G5 are decreased from
100 MW and 61.9 MW in scenario 1 to 76.3 MW and 0 MW at hour 21 in scenario 2, respectively. Here, as
the natural gas demand increases the pressure at junctions will violate the operation limits. The natural gas
pressure at junction J3 will increase to 172.9 psig at hour 21 which violates the upper limit for natural gas
pressure at this junction. The commitment states of the generation units in scenario 1 are given in Table 6-
15. Here the total operation cost is $1,603,326.
To ensure that the unit commitment solution in scenario 1 provides a feasible solution for the
natural gas operation problem, the demands of the natural gas generation units are imposed to the natural
J3
v2
v1
J12
L4
J1
J8
J7
J9
J2J5
J4
L1
G1
J6
L3
G4
G5
L2
G2
G3
J11
v3G6
J10
G7
G8
Gas resource
Compressor
Junction
124
gas network. In this case, the natural gas network failed to serve the demand of the natural gas generation
units. In scenario 2, the operation of the electricity network is coordinated with the operation of the natural
gas network. The changes in the commitment of the generation units compared to those in scenario 1 are
given in Table 6-16. Here, the total operation cost is increased to $1,629,829 which is 1.65% higher than
that in scenario 1. However, unlike scenario 1, the presented solution for unit commitment is feasible for
the natural gas network as there is no supply shortage for the natural gas generation units and there is no
violation of the operation constraints of the natural gas network. Compared to scenario 1, the natural gas
units G3 and G5 are not committed at hours 6-24 and 3-24 in scenario 2, respectively. Therefore, the more
expensive fossil fuel units T13 and T34 are committed during hours 1-22 and 8-13, respectively. Moreover,
G4 is committed at hours 5-13, G2 is committed at hour 4, T14 is decommitted in hour 8 and T46 is
decommitted during hours 1-14 and 21-24.
Table 6-15 Hourly unit commitment in Scenario 1
Unit Hours (1-24)
T1-T3 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
T4-T5 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
T6 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
T7 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
T8-T9 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
T10-T11 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
T12 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 0 0 0
T13 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
T14 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
T15 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
T16 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
T17-T18 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
T19 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
T20-T21 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
T22-T23 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
T24-T25 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
T26 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
T27-T29 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
T30-T31 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
T32 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
T33 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
T34 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1
T35-T36 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
T37 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
T38 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
T39-T40 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
T41-T42 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
T43-T45 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
T46 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 0 0 0
G1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
G2 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
G3 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
G4 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1
G5-G8 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
125
Table 6-16 Changes in the hourly commitment in scenario 2 from scenario 1
Unit Hours (1-24)
T13 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 0 0 0
T14 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
T34 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
T46 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 0 0 0
G2 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
G3 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
G4 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
G5 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
6.4 Summary
This chapter presents a consensus optimization framework for the coordinated operation of electricity and
natural gas networks. Tight convex relaxations for the electricity and natural gas operation problems with
nonconvex feasible regions were proposed. The nonconvex feasible region is characterized by the
Weymouth equation in the natural gas network, and the introduced unit commitment binary decision
variables in the electricity network. A computationally tractable convex representation with several valid
constraints is formulated that procures a rank-1 optimal solution for the natural gas and electricity network
operation problems. The alternating direction method of multipliers is used to solve the convex-relaxed
consensus optimization problem. Two case studies showed the effectiveness of the proposed approach to
capture the interdependence among electricity and natural gas networks.
The presented approach exploits the sparsity of the natural gas network. If the natural gas network
is dense, procuring the rank-1 moment relaxation matrix will be more challenging. Additionally, in order to
reach the global solution, the number of monomials within the vector of monomials could increase
dramatically. Consequently, a very large moment relaxation for each junction of the natural gas network is
formed. Although the presented algorithm would avoid forming the unnecessary large moment relaxation
matrices, very large moment relaxation matrices may be required to procure a rank-1 moment relaxation
matrix. This work could be further extended to capture a more detailed model of the natural gas network
considering the nonlinearities in power consumption of the compressors.
126
Chapter 7
7SUMMARY
The aging electricity grid is the largest machine in the world which is so vulnerable that may fail
under extreme conditions. Although alternating the inexpensive bulk power generation from coal and oil
with the clean energy resources (e.g. natural gas and volatile renewable energy) will reduce the air
pollution and carbon footprint of human activities, a comprehensive economic justification is required to
encourage the investment in grid modernization and clean energy generation. The research presented in this
dissertation employs the cutting-edge mathematical techniques to resolve these key challenges to making
the electricity grid smart.
With the rise in the reliability and resiliency concerns in electricity networks with high penetration
of distributed renewable energy resources, the expansion of microgrids in active distribution network has
become a crucial interest. This problem is analytically addressed by utilizing an advanced graph theory
practice. The expansion of microgrids will facilitate the controllability of distributed energy resources
(DERs) within the distribution network. The proposed framework includes several decision-making
components to identify the most controllable and reliable plan according to attributes and uncertain
parameters including the renewable generation, demand, and component outages. Thus, the planners can
determine the points of common couplings (e.g. boundaries) for microgrids. This mathematical
methodology is a powerful tool that to utilize in undertaking other similar problems in smart grid that
certain level of autonomy in control and privacy is desired while the quality and reliability of the service in
the network are maintained. Another goal that is carried out in the presented dissertation is the
enhancement in the resiliency of energy supply in distribution network through multiple energy carrier
microgrids. Abundance, expanded transportation network, availability at a low price, and minimum CO2
emissions of natural gas prompted the growth in the penetration level of DERs supplied by natural gas. On
the other hand, more than half of the cyber threats occur in the energy sector, so it is pivotal to safeguard
the cyber layer of the cyber-physical system, e.g. the control of multiple energy carrier microgrids. A
mathematical model is developed to identify and then reinforce the vital components within microgrids
127
throughout a preventive reinforcement procedure. In addition, a quantitative resiliency measure is
introduced to analyze the operation of microgrids under threat. There is a tradeoff between minimizing the
investment on the reinforcing the components and maximizing the resiliency of energy supply within the
microgrid. The presented research provides a decision support system and resiliency measure that assists
the microgrid planners to allocate the resources and reach the best strategy based on their budget and their
desired level of resiliency.
Investment on microgrids is unobtainable without a financial model that justifies a decent return
on the investment. Due to the technical limitations and ineligibility of microgrids to directly participate in
the wholesale electricity market, a hierarchical electricity market structure is proposed to facilitate
acquiring profit from microgrids in the grid-connected mode. In the hierarchical market, microgrids
participate in the wholesale electricity market through load aggregators (LAs). A dynamic game theory
approach is employed to model the role of LAs as intermediate agent coupling the wholesale electricity
market with several distribution networks. The proposed hierarchical structure utilized dynamic game
theory is a promising tool to address the interdependence of multiple competitive environments.
Exploring additional aspects of multiple challenges that is addressed in the presented dissertation
requires solving the nonconvex optimization problems. Thus, the convex relaxation techniques is employed
to develop a solid problem-solving tool for overcoming this challenge particularly for smart grid
applications. An algorithm based on sparse semidefinite programming (SDP) relaxation approach is
developed to procuring an at least near-global optimal solution of these problems. By leveraging the
sparseness of the networks, the presented algorithm construct several small, tight, and traceable SDP
relaxation matrices. This technique opened a new door to solve several challenging problems including the
non-convexity issue within the natural gas network.
128
APPENDIX
The research conducted to address the problems discussed in this dissertation lead solution methodology to
overcome a wide-range of challenges in grid modernization. It should be noted that only a selected number
of works are discussed in this dissertation. The relevant publications are listed as followings.
- S. D. Manshadi and M. E. Khodayar. Coordinated Operation of Electricity and Natural Gas
Systems: A Convex Relaxation Approach. IEEE Transactions on Smart Grid, accepted for
publication, 2018.
- S. D. Manshadi and M. E. Khodayar. A Tight Convex Relaxation for the Natural Gas Operation
Problem. IEEE Transactions on Smart Grid, accepted for publication, 2018.
- S. D. Manshadi and M. E. Khodayar. Risk-Averse Generation Maintenance Scheduling with
Microgrid Aggregators. IEEE Transactions on Smart Grid, in press, 2017.
- S. D. Manshadi and M. E. Khodayar. Wireless Charging of Electric Vehicles in Electricity and
Transportation Networks. IEEE Transactions on Smart Grid, in press, 2017.
- A. Vafamehr, M. E. Khodayar, S. D. Manshadi, I. Ahmed, adn J. Lin. A Framework for
Expansion Planning of Data Centers in Electricity and Data Networks under Uncertainty. IEEE
Trans. on Smart Grid, in press, 2017.
- S. D. Manshadi and M. E. Khodyar. Expansion of Autonomous Microgrids in Active Distribution
Networks. IEEE Transactions on Smart Grid, in press, 2016.
- M. E. Khodayar, S. D. Manshadi, and A. Vafamehr. Short-term Operation of Microgrids in
Transactive Energy Architecture. The Electricity Journal, 29(10): 41–48, Dec. 2016.
- M. E. Khodayar, S. D. Manshadi, H. W, and J. Lin. Multiple Period Ramping Processes in Day-
Ahead Electricity Markets. IEEE Transactions on Sustainable Energy, 7(4):1634–1645, Oct. 2016.
- S. D. Manshadi and M. E. Khodyar. A Hierarchical Electricity Market Structure for the Smart
Grid Paradigm. IEEE Transactions on Smart Grid, 7(4):1866–1875, Jul. 2016.
129
- S. D. Manshadi and M. E. Khodyar. Resilient Operation of Multiple Energy Carrier Microgrids.
IEEE Transactions on Smart Grid, 6(2):2283–2392, Sep. 2015. UNDER REVIEW
- S. D. Manshadi, G. Liu, M. E. Khodayar, J. Wang, and R. Dai. A Convex Relaxation Approach to
Find a Feasible Solution for the Power Flow Problem. IEEE Transactions on Power Systems,
Under Review.
- S. D. Manshadi and M. E. Khodayar. Strategic Behavior of In-motion Wireless Charging
Aggregators in the Electricity and Transportation Networks. IEEE Transactions on Smart Grid,
Under Review.
- S. D. Manshadi, M. E. Khodayar, and C. Chen. Topology Estimation in Mesh Distribution
Network with Limited Measurements. IEEE Transactions on Smart Grid, Under Review.
- S. D. Manshadi and M. E. Khodayar. Decentralized Operation Framework for Hybrid AC/DC
Microgrids. Proceedings of the North America Power Symposium (NAPS 2016), Sep. 2016.
130
BIBLIOGRAPHY
[DOE10]
U.S. Department of Energy, Hardening and Resiliency: U.S. Energy Industry Response
to Recent Hurricane Seasons, Tech. Rep., Aug. 2010 [Online]. Available:
http://www.oe.netl.doe.gov/docs/HR-Report-final-081710.pdf
[Abd15]
A. Abdulwahab, A. Abusorrah, X. Zhang, and M. Shahidehpour, “Coordination of
interdependent natural gas and electricity infrastructures for firming the variability of
wind energy in stochastic day-ahead scheduling,” IEEE Trans. Sust. Energy, vol. 6, no.
2, pp. 606-615, Apr. 2015.
[Ach09]
T. Achterberg, “SCIP: Solving Constraint Integer Programs,” Mathematical
Programming Computation, vol. 1, no.1, pp. 1-41, Jul. 2009.
[Act15]
Fairfield: A Connecticut Town on the Vanguard of Microgrid Development, Schneider
Electric, Fairfield Case Study, Jan. 2015 [Online]. Available: http://se-enable.com/wp-
content/uploads/MG_Fairfield-Case-Study_Digital.pdf
[Akb11]
A. Akbari Goroud, M. Amirahmadi, M. Bahmanzadeh, and A. Akbar Abdoos, “Optimal
bidding strategy for all market players in a wholesale power market considering
demand response programs”, European Trans on Elec. Power, vol. 21, no 1, pp. 293-
311, 2011.
[Alg09]
A. A. S. Algarni and K. Bhattacharya, “A generic operations frame work for discos in
retail electricity market,” IEEE Trans. Power Syst., vol. 24, no. 1, pp. 356–367, Feb.
2009.
[Alm]
M. Almassalkhi and I. Hiskens, “Optimization framework for the analysis of large-scale
networks of energy hubs,” in Proc. 17th Power Syst .Computation Conf., Sweden, 2011.
[Arr10] J M. Arroyo, “Bilevel programming applied to power system vulnerability analysis
131
under multiple contingencies,” IET Gen. Trans. Dist., vol. 4, no. 2, pp. 178–190, Feb.
2010.
[Asi13]
G. E. Asimakopoulou, A. L. Dimeas, N. D. Hatziargyriou, “Leader-follower strategies
for energy management of multi-microgrids,” IEEE Trans. Smart Grid, vol. 4, no. 4, pp.
1909–1916, Dec. 2013.
[Awa09]
B. Awad, J. Wu, J. Ekanayake, and N. Jenkins, “Integrated optimal power flow for
electric power and heat in a MicroGrid,”– Part1, 20th Intenational Conf. and Exhib.
Elec. Dist, Czech Repubic, 2009.
[Bah16]
S. Bahram and A. Sheikhi, “From demand response in smart grid toward integrated
demand response in smart energy hub, IEEE Trans. Smart Grid, vol. 7, no. 2, pp. 650-
658, 2016.
[Bar98]
J. Bard, Practical Bilevel Optimization: Algorithms and Applications. Norwell, MA,
USA: Kluwer, 1998.
[Bas10]
A. K. Basu, S. Chowdhury, and S. P. Chowdhury, ‘‘Distributed energy resource
capacity adequacy assessment for PQR enhancement of CHP micro-grid,’’ in Proc.
IEEE PES General Meeting, Jul. 2010, pp. 1–5.
[Bas95]
T. Basar and G. J. Olsder, ¸ Dynamic Non-Cooperative Game Theory, 2nd Ed. New
York: Academic, 1995.
[Ben09]
A. Ben-Tal, G. Boaz, S. Shimrit. “Robust multi-echelon multi-period inventory
control,” European Journal of Operational Research, vol. 199, no. 3, pp. 922-935,
2009.
[Ber11]
D. Bertsimas, D. B. Brown, C. Caramanis, Theory and applications of robust
optimization. SIAM Review vol. 53, no. 3, pp. 464-501, 2011.
[Ber13]
D. Bertsimas, E. Litvinov, X. A. Sun, J. Zhao, and T. Zheng, “Adaptive robust
optimization for the security constrained unit commitment problem,” IEEE Trans.
Power Syst., vol. 28, no. 1, pp. 52-63, Feb. 2013.
[Bie07] D. Bienstock and S. Mattia, “Using mixed-integer programming to solve power grid
132
blackout problems,” Discrete Optimization, vol. 4, pp. 115–141, 2007.
[Bil96]
R. Billinton and R. Allan, Reliability Evaluation of Power Systems, 2nd ed. London,
U.K.: Plenum, 1996.
[Bor12]
A. Borghetti, “A mixed-integer linear programming approach for the computation of the
minimum-losses radial configuration of electrical distribution networks,” IEEE Trans.
Power Syst., vol. 27, no. 3, pp.1264 -1273, 2012.
[Boy11]
S. Boyd, N. Parikh, E. Chu, B. Peleato, and J. Eckstein, “Distributed optimization and
statistical learning via the alternating direction method of multipliers,” Foundations and
Trends R in Machine Learning, vol. 3, no. 1, pp. 1–122, 2011.
[Boy93]
S. P. Boyd and L. El Ghaoui, “Method of centers for minimizing generalized
eigenvalues,” Lin. Alg. Appl., vol. 188, pp. 63–111, 1993.
[Bus14]
M. R. Bussieck and S. Vigerske, “MINLP solver software,” [Online]. Available:
http://www.math.hu-berlin.de/~stefan/minlpsoft.pdf
[Cgf]
Compressible gas flow, Pipe Flow Calculations. [Online]. Available:
http://www.pipeflowcalculations.com/pipe-valve-fitting-flow/compressible-gas-
flow.php
[Cor14]
C. M. Correa-Posada and P. Sanchez-Martin, “Security-constrained optimal power and
natural-gas flow,” IEEE Trans. Power Syst., vol. 29, no. 4, pp. 1780-1787, Jul. 2014.
[Dae02] J. Daemen and V. Rijmen, The Design of Rijndael. Springer-Verlag, 2002.
[DeM09]
V. DeMiguel and H. Xu. “A stochastic multiple-leader stackelberg model: Analysis,
computation, and application,” Oper. Res., vol. 57, no. 5, pp. 1220-1235, Sep. 2009.
[Den82] D.E. Denning, Cryptography and Data Security. Addison-Wesley, 1982.
[Doo12]
M. Doostizadeh and H. Ghasemi, “Day-ahead scheduling of an active distribution
network considering energy and reserve markets,” European Trans. Electric Power,
vol. 19, doi: 10.1002/etep.1630 2012.
[Dup03]
J. Dupacová, N. Gröwe-Kuska, and W. Römisch, “Scenario reduction in stochastic
programming: An approach using probability metrics,” Math. Program., vol. 95, pp.
133
493–511, 2003.
[Fie73]
M. Fiedler, “Algebraic connectivity of graphs,” Czech. Math. Journal, vol. 23, no. 98,
pp. 298-305, 1973.
[Fie75]
M. Fiedler,”A property of eigenvectors of nonnegative symmetric matrices and its
application to graph theory,” Czech. Math. Journal, vol. 25, no. 4, pp. 619-633, 1975
[Fle04]
R. Fletcher and S. Leyffer, “Solving mathematical programs with complementarity
constraints as nonlinear programs,” Optimiz. Meth. Softw., vol. 19, pp. 15–40, Feb.
2004.
[FMO15]
Forward Market Operations. Energy & Ancillary Services Market Operations, M-11
Rev. 75. Technical report, PJM, 2015.
[Fu07]
Y. Fu, M. Shahidehpour, and Z. Li, “Security-constrained optimal coordination of
generation and transmission maintenance outage scheduling,” in IEEE Trans. Power
Systems, vol. 22, no. 3, pp. 1302-1313, Aug. 2007.
[Fu09]
Y. Fu, Z. Li, M. Shahidehpour, T. Zheng and E. Litvinov, “Coordination of midterm
outage scheduling with short-term security-constrained unit commitment,” in IEEE
Trans. Power Syst., vol. 24, no. 4, pp. 1818-1830, Nov. 2009.
[Gab12]
S. A. Gabriel, A. J. Conejo, J. D. Fuller, B. F. Hobbs, and C. Ruiz, Complementarity
Modeling in Energy Markets. Berlin, Germany: Springer, 2012.
[GAM]
GAMS/SCENRED Documentation [Online]. Available:
http://www.gams.com/docs/document.htm.
[Gao16]
H. Gao, Y. Chen, Y. Xu, and C.-C. Liu, “Resilience-oriented critical load restoration
using microgrids in distribution systems” IEEE Trans. Smart Grid, DOI:
10.1109/TSG.2016.2550625
[Gei06]
M. Geidl and G. Andersson,”Operational and structural optimization of multi-carrier
energy systems,” European Transactions on Electrical Power, 16(5):463-477, 2006.
[Gei07]
M. Geidl and G. Andersson, “Optimal power flow of multiple energy carriers,” IEEE
Trans. Power Syst., vol. 22, no. 1, pp. 145-155, 2007.
134
[Gka13]
L. Gkatzikis, I. Koutsopoulos, and T. Salonidis, “The role of aggregators in smart grid
demand response markets,” IEEE Journal Selected Areas in Comm., vo. 31, no. 7,
pp.1247-1257, Jul. 2013.
[Gre03]
W. H. Greene, Econometric Analysis, 5th ed. Upper Saddle River, NJ: Prentice-Hall,
2003.
[Hag12]
H. Haghighat and S. W. Kennedy, “A bilevel approach to operational decision making
of a distribution company in competitive environments,” IEEE Trans. Power Syst., vol.
27, no. 4, pp. 1797–1807, Nov. 2012.
[He17]
C. He, L. Wu, T. Liu, and M. Shahidehpour, “Robust co-optimization scheduling of
electricity and natural gas systems via admm,” IEEE Trans. Sustain. Ener., vol. pp, no.
99, pp. 1–12.
[Hib12]
P. J. Hibbard and T. Schatzki, “The interdependence of electricity and natural gas:
Current factors and future prospects,” Electricity J., vol. 25, no. 4, pp. 6–17, May 2012.
[Iag13]
"IEEE application guide for IEEE Std 1547, 1547.4-2011 - IEEE Guide for Design,
Operation, and Integration of Distributed Resource Island Systems with Electric Power
Systems " in IEEE Std 1547.4-2011, 2011, pp. 1-54.
[ICS13]
ICS-CERT incident response summary report. 2013. Industrial Control Systems Cyber
Emergency Response Team (ICS-CERT). https://ics-cert.us-
cert.gov/sites/default/files/ICS-CERT_Monitor_April-June2013.pdf
[ICS14]
ICS-CERT incident response summary report. 2014. Industrial Control Systems Cyber
Emergency Response Team (ICS-CERT). https://ics-cert.us-
cert.gov/sites/default/files/Monitors/ICS-CERT_Monitor_ Jan-April2014.pdf
[1]
[Lof04]
[2] J. Lofberg, “Yalmip: A toolbox for modeling and optimization in MATLAB,” in Proc.
IEEE Int. Symp. Comput. Aided Control Syst. Des., Taipei, Taiwan, Sep. 2004, pp. 284–
289.
[IEEE123]
IEEE 123 node test feeder, Distribution system analysis subcommittee. [Online].
Available: http://ewh.ieee.org/soc/pes/dsacom/testfeeders/
[Imi15] Innovative microgrid improves utility’s reliability and optimizes distributed energy
135
resources, Schneider Electric, Oncor Case Study, Apr. 2015 [Online]. Available:
http://microgrids.schneider-electric.us/wp-content/uploads/Oncor-case-study-vF-
4.13.15-2.pdf
[Ji16]
G. Ji, W. Wu, and B. Zhang, "Robust generation maintenance scheduling considering
wind power and forced outages," IET Renewable Power Generation, vol. 10, no. 5, pp.
634-641, May 2016.
[Kar14]
A. Kargarian, Y. fu, S. DorMohammadi, M. Rais-Rohani, “Optimal operation of active
distribution grids: a system of systems framework,” IEEE Trans. Smart Grid, vol. 5, no.
3, pp. 1228–1237, May. 2014.
[Kho12]
M. E. Khodayar, M. Barati, and M. Shahidehpour, “Integration of high reliability
distribution system in microgrid operation,” IEEE Trans. Smart Grid, vol. 3, no. 4, pp.
1997–2006, Dec. 2012.
[Kho14]
M. E. Khodayar, and M Shahidehpour, “Optimal strategies for multiple participants in
electricity markets ,” IEEE Trans. Power Syst., vol. 29, no. 2, pp. 986–987, Mar. 2014.
[Kim11]
H. Kim and M. Thottan, “A two-stage market model for microgrid power transactions
via aggregators,” Bell Labs Technical Journal, vol. 16, no. 3, pp. 101–107, Nov. 2011.
[Koc16]
B. Kocuk, S. Dey, and A. Sun, “Strong SOCP relaxation for the optimal power flow
problem,” Oper. Res. Vol. 64, no. 6, pp. 1177-1196, Dec. 2016.
[Koe04]
G. Koeppel, M. Geidl, G. Andersson, “Value of storage devices in congestion
constrained distribution networks,” in Proc. Power Syst. Technology, PowerCon 2004,
vol. 1, pp. 624-629, Nov. 2004.
[Las10]
J.-B. Lasserre, Moments, Positive Polynomials and Their Applications. Imperial
College Press, 2010.
[Las11]
R. H. Lasseter, “Smart distribution: coupled microgrids,” in Proc. of the IEEE, vol. 99,
no. 6, pp. 1074-1082, Jun. 2011.
[Ley01]
S. Leyffer, “Integrating SQP and branch-and-bound for mixed integer nonlinear
programming,” Computational Optimization and Applications, vol. 18, no. 3, pp. 295-
136
309, 2001.
[Ley10]
S. Leyffer and T. Munson, “Solving multi-leader-common-follower games,” Optimiz.
Meth. Softw., vol. 25, pp. 601–623, Aug. 2010.
[Li05]
T. Li and M. Shahidehpour, “Strategic bidding of transmission constrained GENCOs
with incomplete
[Li07]
H. Li, Y. Li, and Z. Li, “A multi-period energy acquisition model for a distribution
company with distributed generation and interruptible load,” IEEE Trans. Power Syst.,
vol. 22, no. 2, pp. 588–596, May 2007.
[Li08]
T. Li, M. Eremia, and M. Shahidehpour, “Interdependency of natural gas network and
power system security,” IEEE Trans. Power Syst., vol. 23, no. 4, pp. 1817-1824, Nov.
2008.
[Li14]
J. Li, X.-Y. Ma, C.-C. Liu, and K. P. Schneider, “Distribution system restoration with
microgrids using spanning tree search,” IEEE Trans. Power Syst., vol. 29, no. 6, pp.
3021-3029, Nov. 2014.
[Liu09]
C. Liu, M. Shahidehpour, Y. Fu and Z. Li, "Security-Constrained Unit Commitment
With Natural Gas Transmission Constraints," IEEE Trans. Power Systems,, vol. 24, pp.
1523-1536, 2009.
[Liu11]
C. Liu, M. Shahidehpour, and J. Wang, “Coordinated scheduling of electricity and
natural gas infrastructures with a transient model for natural gas flow,” Chaos J., vol.
21, Jun. 2011.
[Lop11]
J. M. López-Lezama, A. Padilha-Feltrin, J. Contreras, and J. I. Muñoz, “Optimal
contract pricing of distributed generation in distribution networks,” IEEE Trans. Power
Syst., vol. 26, no. 1, pp. 128–136, Feb. 2011.
[Man16]
S. D. Manshadi and M.E. Khodayar, “A hierarchical electricity market for the smart
grid paradigm,” IEEE Trans. Smart Grid, vol. 7, no. 4, pp. 1866-1875, Jul. 2016.
[Mar12]
A. Martinez-Mares and F. R. Claudio, "A unified gas and power flow analysis in natural
gas and electricity coupled networks." IEEE Trans. Power Syst. vol 27, no. 4, pp. 2156-
137
2166, 2012.
[Mar13]
A. Martinez-Mares and F. R. Claudio, "A robust optimization approach for the
interdependency analysis of integrated energy systems considering wind power
uncertainty," IEEE Trans. Power Syst., vol. 28, no. 4, pp. 3964-3976, 2013.
[Mar16]
G. Maria, P. Duenas, and J. Reneses, “Electricity and natural gas interdependency:
comparison of two methodologies for coupling large market models within the
European regulatory framework," IEEE Trans. Power Syst. vol. 31, no. 1, pp. 361-369,
2016.
[Mas11]
M. Mashhour, M. A. Golkar, S. M. Moghaddas-Tafreshi, “Extending market activities
for a distribution company in hourly-ahead energy and reserve markets – Part I:
Problem formulation,” Energy Conv. and Manage, vol. 52, no. 1, pp.477-486, Jan.
2011.
[Men05] E. S. Menon, Gas Pipeline Hydraulics. New York: Taylor & Francis, 2005.
[Mic] Microgrid definitions, [online] https://building-microgrid.lbl.gov/microgrid-definitions
[Mir05]
I. Mironov, "Hash functions: Theory, attacks, and applications," [Online], Microsoft
Research, Silicon Valley Campus, Nov. 2005.
[Moe14]
M. Moeini-Aghtaie, A. Abbaspour, M. Fotuhi-Firuzabad, and E. Hajipour. "A
decomposed solution to multiple-energy carriers optimal power flow." IEEE Trans on
Power Syst, vol. 29, no. 2, pp. 707-716, 2014.
[Mol15]
D. K. Molzahn and I. A. Hiskens, “Sparsity-exploiting moment-based relaxations of the
optimal power flow problem,” IEEE Trans. Power Syst., vol. 30, no. 6, pp. 3168-3180,
Nov. 2015.
[Mol17]
M. Mollahassani-pour; M. Rashidinejad; A. Abdollahi; M. A. Forghani, “Demand
response resources’ allocation in security-constrained preventive maintenance
scheduling via MODM method,” in IEEE Syst. Journal, vol. PP, no. 99, pp.1-12.
[Mot05]
A. L. Motto, J.M. Arroyo, and F.D. Galiana, “A mixed-integer LP procedure for the
analysis of electric grid security under disruptive threat”, IEEE Trans. Power Syst., vol.
138
20, no. 3, pp. 1357-1365, Aug 2005.
[Nak13]
K. Nakayama, N. Shinomiya, and H. Watanabe, “An autonomous distributed control
method based on tie-set graph theory in future grid", Inter. Jour. of Cir. Th. and
App., Volume.41, Issue.11, pp.1154, 2013.
[Nii00]
S. Niioka, N. Okada, and R. Yokoyama, “Evaluation and allocation of supply reliability
cost in electricity market,” in Proc. PowerCon 2000, Dec. 2000, vol. 2, 4–7, pp. 703–
707.
[Ozt05]
U. A. Ozturk, M. Mazumdar, B. A. Norman. “A solution to the stochastic unit
commitment problem using chance-constrained programming,” IEEE Trans. Power
Systems, vol. 19, no. 3, pp. 1589-1598, Aug. 2004.
[Pal05]
R. Palma-Behnke, J. L. A. Cerda, L. Vargas, and A. Jofre, “A distribution company
energy acquisition market model with the integration of distribution generation and load
curtailment options,” IEEE Trans. Power Syst., vol. 20, no. 4, pp. 1718–1727, Nov.
2005.
[Par13]
M. Parvania, M. Fotuhi-Firuzabad, and M. Shahidehpour, “Optimal demand response
aggregation in wholesale electricity markets,” IEEE Trans. Smart Grid, vol. 4, no. 4,
pp. 1957–1964, Dec. 2013.
[Pra85]
J. F. Prada, The Value of Reliability in Power Systems, Pricing Operating Reserves,
MIT EL-99-005 WP, Energy Laboratory, MIT, Jun. 1985.
[Qad14]
M. Qadrdan, J. Wu, N. Jenkins, and J. Ekanayake, “Operating strategies for a GB
integrated gas and electricity network considering the uncertainty in wind power
forecasts,” IEEE Trans. Sustain. Energy, vol. 5, no. 1, pp. 128-138. Jan. 2014.
[Que07]
A. Quelhas, E. Gil, J. McCalley, S. M. Ryan, “Multiperiod Generalized Network Flow
Model of the U.S. Integrated Energy System: Part I—Model Description,” IEEE Trans.
Power Syst., vol. 22, no. 2, pp. 829-836, May 2007.
[Rah07]
S. Rahman, M. Pipattanasomporn, and Y. Teklu, “Intelligent distributed autonomous
power system (IDAPS),” in Proc. IEEE PES Gen. Meet., 2007, pp. 1–8.
139
[Rin04]
S. M. Rinaldi, "Modeling and Simulating Critical Infrastructures and Their
Interdependencies," in Proc. of the 37th Annual Hawaii International Conference on
System Sciences (HICSS'04), (Big Island, HI, USA), IEEE Computer Society Press, Jan.
2004.
[Rom12]
N. Romero, N. Xu, L. K. Nozick, I. Dobson, and D. Jones, “Investment planning for
electric power systems under terrorist threat,” IEEE Trans. Power Syst., vol. 27, no. 1,
pp. 108–116, Feb. 2012.
[Rom97] G. Romp, Game Theory: Introduction and Applications, Oxford University Press, 1997.
[Roy00]
S. Roy, and B. Lesieutre, “Studies in network partitioning based on topological
structure,” 32nd Annual North American Power Symposium, Waterloo, Canada,
Oct.2000.
[Rui09]
P. A. Ruiz, C. R. Philbrick, E. Zak, K. W. Cheung, P. W. Sauer, “Uncertainty
management in the unit commitment problem,” IEEE Transactions on Power Systems,
vol. 24, no. 2, pp. 642-651, May 2009.
[Rui12]
C. Ruiz, A. J. Conejo, and Y. Smeers, “Equilibria in an oligopolistic electricity pool
with stepwise offer curves,” IEEE Trans. Power Syst., vol. 27, no. 2, pp. 752–761, May
2012.
[Saa12]
W. Saad, Z. Han, H. Poor, and T. Basar, “Game-Theoretic Methods for the Smart Grid:
An Overview of Microgrid Systems, Demand-Side Management, and Smart Grid
Communications,” IEEE Signal Process. Mag., vol. 29, pp. 86–105, Sep. 2012.
[Saf13]
A. Safdarian, M. Fotuhi-Firuzabad, and M. Lehtonen, “A stochastic framework for
short-term operation of a distribution company,” IEEE Trans. Power Syst., vol. 28, no.
4, pp. 4712–4721, Nov. 2013.
[Sal04]
J. Salmeron, K. Wood, and R. Baldick, “Analysis of electric grid security under terrorist
threat,” IEEE Trans. Power Syst., vol. 19, no. 2, pp. 905–912, May 2004.
[Sal13]
C. A. Saldarriaga, R. A. Hincapie, and H. Salazar, “A Holistic Approach for Planning
Natural Gas and Electricity Distribution Networks,” IEEE Trans. Power Syst., vol. 28,
140
no. 4, pp. 4052-4063, Nov. 2013.
[San16]
C. B. Sanchez, R. Bent, S. Backhaus, S. Blumsack, H. Hijazi, and P. van Hentenryck,
“Convex optimization for joint expansion planning of natural gas and power systems,”
in 49th Hawai’i Int’l. Conf. Sys. Sci., pp. 1–10, Jan 2016.
[Sha02]
M. Shahidehpour, H. Yamin, and Z. Li, Market Operations in Electric Power Systems:
Forecasting, Scheduling, and Risk Management. New York, NY, USA: Wiley, 2002.
[Sha05]
M. Shahidehpour, Y. Fu, and T. Wiedman, “Impact of natural gas infrastructure on
electric power systems,” Proc. IEEE, vol. 93, no. 5, pp. 1042–1056, May 2005.
[Sha12]
M. H. Shariatkhah, M. R. Haghifam, “Using feeder reconfiguration for congestion
management of smart distribution network with high DG penetration,” in Proc.
Integration of Renewables into Distribution Grid, pp. 1-4, May 2012.
[Sha14]
M. Shahidehpour and Z. Li. White paper: Long-term electric and natural gas
infrastructure requirements. Technical report, Illinois Institute of Technology, 2014.
[Sha17]
C. Shao, X. Wang, M. Shahidehpour, X. Wang, and B. Wang, “An MILP-Based
Optimal Power Flow in Multicarrier Energy Systems,” IEEE Trans. Sust. Energy, vol.
8, no.1, pp. 239-248, Jan. 2017.
[Sha15]
S. Shafiq, A. T. Al-Awami, “Reliability and economic assessment of renewable
microgrid with V2G electric vehicles coordination,” IEEE Jordon. Conf. App. Elec.
Eng. And Comp, Tech, 2015, pp. 1-6.
[She84]
H. D. Sherali, “A multiple leader Stackelberg model and analysis,” Oper. Res., vol. 32,
pp. 390–404, Mar. 1984.
[Taw04]
M. Tawarmalani and N.V. Sahinidis,”Global optimization of mixed-integer nonlinear
programs: A theoretical and computational study,” Mathematical Programming,” vol.
99, no. 3, pp. 563-591, Apr. 2004.
[Tep12]
Terrorism and the Electric Power Delivery System,” Nat. Acad. Science, Nat. Res.
Council study rep., 2012.
[Tom07] A. Tomasgard, F. Rømo, M. Fodstad, and K. Midthun, “Optimization models for the
141
natural gas value chain,” Geom. Model., Numer. Simul., Optimiz., pp. 521-558, 2007.
[Tou09]
A. Tuohy, P. Meibom, E. Denny, M. O’Malley, “Unit commitment for systems with
significant wind penetration,” IEEE Transactions on Power Systems, vol. 24, no. 2, pp.
592-601, May 2009.
[Tra11]
T. Krause, G. Andersson, K. Frohlich and A. Vaccaro, "Multiple-energy carriers:
modeling of production, delivery, and consumption", Proc. IEEE, vol. 99, no.
1, pp.15 -27 2011.
[Uns10]
C. Unsihuay-Vila, J. W. Marangon-Lima, A. C. Zambroni de Souza, I. J. Perez-Arriaga,
and P. P. Balestrassi,, “A model to long-term, multiarea, multistage, and integrated
expansion planning of electricity and natural gas systems,” IEEE Trans. Power Syst.,
vol. 25, no. 2, pp. 1154–1168, May 2010.
[Vig13]
S. Vigerske, “Decomposition of Multistage Stochastic Programs and a Constraint
Integer Programming Approach to Mixed-Integer Nonlinear Programming,” PhD
thesis, Humboldt Universitat zu Berlin, 2013.
[Wan08]
J. Wang, M. Shahidehpour, Z. Li. “Security-constrained unit commitment with volatile
wind power generation,” IEEE Transactions on Power Systems, vol. 23, no. 3, pp.
1319-1327, Aug. 2008.
[Wan12]
Q. Wang, Y. Guan, J. Wang, “A chance-constrained two-stage stochastic program for
unit commitment with uncertain wind power output,” IEEE Transactions on Power
Systems, vol. 27, no. 1, pp. 206-215, Feb. 2012.
[Wan15]
Z. Wang, B. Chen, J Wang, and M.M Begovic,”Coordinated energy management of
networked microgrids in distribution systems,” IEEE Trans. Smart Grid, vol. 6, no. 1,
pp. 44-53, Jan. 2015.
[Wan16a]
Y. Wang; Z. Li; M. Shahidehpour; L. Wu; C. X. Guo; B. Zhu, “Stochastic Co-
Optimization of Midterm and Short-Term Maintenance Outage Scheduling Considering
Covariates in Power Systems,” in IEEE Trans. Power Syst., vol. pp, no.99, pp.1-11,
2016.
142
[Wan16b]
Y. Wang, H. Zhong, Q. Xia, D. S. Kirschen and C. Kang, “An approach for integrated
generation and transmission maintenance scheduling considering N-1 contingencies,”
IEEE Trans. on Power Syst., vol. 31, no. 3, pp. 2225-2233, May 2016.
[Wen17]
Y. Wen, X. Qu, W. Li, X. Liu, X. Ye, “Synergistic Operation of Electricity and Natural
Gas Networks via ADMM,” IEEE Trans. Smart Grid, vol. PP, no.99, pp.1-1, 2017.
[Wes02]
T. Westerlund and R. Porn, “Solving pseudo-convex mixed integer optimization
problems by cutting plane techniques,” Optimization and Engineering, vol. 3, no. 3, pp.
253-280, Sep 2002.
[Wes03]
T. Westerlund and K. Lundquist, “Alpha-ECP, version 5.04. an interactive
MINLPsolver based on the extended cutting plane method,” Technical Report 01-178-
A, Process Design Laboratory, Abo Akademi University, Abo, Finland, 2003. [Online].
Available: http://www.abo.fi/~twesterl/A-ECPManual.pdf.
[Woo96]
A. J. Wood and B. F. Wollenberg, Power Generation, Operation and Control, 2nd Ed.
New York: Wiley, 1996.
[Wu07]
L. Wu, M. Shahidehpour, and T. Li. “Stochastic security-constrained unit
commitment,” IEEE Transactions on Power Systems, vol. 22, no. 2, pp. 800-811, May
2007.
[Wu08a]
L. Wu, M. Shahidehpour, and Z. Li, “GENCO’s risk-constrained hydrothermal
scheduling,” IEEE Trans. Power Syst., vol. 23, no. 4, pp. 1847–1858, Nov. 2008.
[Wu08b]
L. Wu, M. Shahidehpour and T. Li, “GENCO's Risk-Based Maintenance Outage
Scheduling,” in IEEE Transactions on Power Systems, vol. 23, no. 1, pp. 127-136, Feb.
2008.
[Wu10]
L. Wu, M. Shahidehpour, and Y. Fu, “Security-Constrained Generation and
Transmission Outage Scheduling with Uncertainties,” in IEEE Trans. Power Syst., vol.
25, no. 3, pp. 1674-1685, Aug. 2010.
[Wu17]
F. Wu, H. Nagarajan, A. Zlotnik, and R. Sioshansi, “adaptive convex relaxations for gas
pipeline network optimization,” in proc. American Control Conference (ACC). IEEE,
143
2017.
[Yao07]
Yiming Yao; Edmunds, T.; Papageorgiou, D.; Alvarez, R., "Trilevel Optimization in
Power Network Defense," Systems, Man, and Cybernetics, Part C: Applications and
Reviews, IEEE Transactions on , vol.37, no.4, pp.712,718, July 2007
[Zen13]
B. Zeng and L. Zhao, “Solving two-stage robust optimization problems using a column-
and-constraint generation method,” Oper. Res. Lett., vol. 41, no. 5, pp. 457–461, Sep.
2013.
[Zha14]
L. Zhang, Z. Li, and C.Wu, ”Randomized auction design for electricity markets
between grids and microgrids,” in Proc. ACM SIGMETRICS 2014, Austin, Texas, USA,
Jun. 16-20, 2014.
[Zim09]
R. Zimmerman and C. E. Restrepo, "Analyzing cascading effects within infrastructure
sectors for consequence reduction," in Technologies for Homeland Security, IEEE
Conference On, 2009, pp. 165-170.
[Zlo17]
A. Zlotnik, L. Roald, S. Backhaus, M. Chertkov, and G. Andersson, “Coordinated
scheduling for interdependent electric power and natural gas infrastructures,” IEEE
Trans. on Power Systems, vol. 32, no. 1, pp. 600–610, 2017.
[Zug13]
M. Zugno, J. M. Morales, P. Pinson, and H. Madsen, “A bilevel model for electricity
retailers' participation in a demand response market environment,” Energ. Econ., vol.
36, pp. 182-197, 2013.