06311450 (1)
TRANSCRIPT
-
8/18/2019 06311450 (1)
1/9
1226 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 28, NO. 2, MAY 2013
Detection of High Impedance Faultin Power Distribution Systems Using
Mathematical MorphologySuresh Gautam , Student Member, IEEE , and Sukumar M. Brahma , Senior Member, IEEE
Abstract— A high impedance fault (HIF) is characterized bya small, nonlinear, random, unstable, and widely varying faultcurrent in a power distribution system. HIFs draw very low fault
currents, and hence are not always effectively cleared by conven-
tional overcurrent relays. Various schemes are proposed to detectsuch faults. This paper presents a method to detect HIFs usinga tool based on mathematical morphology (MM). The methodis implemented alongside the conventional overcurrent relay atthe substation to improve the performance of this relay in de-
tecting HIFs. It is rigorously tested on standard test systems using
PSCAD/EMTDC® to generate test waveforms, and Matlab® toimplement the method. Simulation results show that the proposedmethod is fast, secure, and dependable.
IndexTerms— High impedance fault, mathematical morphology,power distribution system, power system protection.
I. I NTRODUCTION
H IGH impedance faults (HIFs) cannot be detected or cleared by conventional overcurrent relays due to verylow fault current. Such faults occur either when a tree limb or
other high impedance objects make contact with the primarydistribution conductor and the ground, or when a conductor
breaks and touches the earth’s surface such as asphalt, con-
crete, grass, sod, sand, etc. These surfaces impose very high
impedances and limit the fault currents to very low values. HIF
studies are reported by the Power System Relaying Committee
(PSRC) working group, which indicates less than 20% success
rate using conventional protection schemes [1]. The undetected
HIFs are hazardous to the public as they leave an energized
conductor exposed and uncleared.
HIF has been a topic of interest since the 1970s with research
focused on investigation of distinguishing characteristics in cur-
rent and voltage waveforms [1]. HIFs typically occur at voltage
levels of 15 kV and below [1], [2]. They are characterized, in
addition to low values of fault currents, by “random behavior
with unstable and wide fluctuation in current” [1]. HIF currents
are found to be composed of harmonics and high frequency
components. During the initial phase, the researchers were pri-
marily engaged with laboratory models and staged fault studies.
Manuscript received January 18, 2012; revised June 01, 2012; accepted Au-gust 14, 2012. Date of publication September 24, 2012; date of current versionApril 18, 2013. Paper no. TPWRS-00056-2012.
The authors are with Klipsch School of Electrical and Computer Engi-neering, New Mexico State University, Las Cruces, NM 88003 USA (e-mail:[email protected]; [email protected]).
Digital Object Identifier 10.1109/TPWRS.2012.2215630
However, with a better understanding of the nature and charac-
teristics of HIF and the advancements in simulation software,
the trend has shifted more towards simulation studies. The HIF
models used for simulation studies are discussed in Section II.
Researchers, from both academia and utilities, have proposed
different methods comprising of single or multiple algorithms
to detect HIFs in a distribution system. Majority of the studies
in this area were reported in 1980s and 1990s, but the sim-
ulation and detection methods are still being developed andrefined. An extensive literature sur vey of HIF can be found in
[3]. The HIF detection methods can be broadly classified into
frequency domain algorithms, time domain algorithms, hybrid
algorithms, and expert systems. Statistical techniques using
sequence components and their harmonics [4], third harmonics
[5], second-fourth-sixth harmonics [6], and high frequency
components (2–10 kHZ) [7] are some of the methods based
on the frequency domain. Other frequency domain algorithms
include the use of burst noise signal [8], improvements over the
use of third harmonic based algorithms [9], [10] and a combi-
nation of current magnitude with the magnitude and phase of its
harmonics [11]. Time domain algorithms include proportionalrelaying [12], ratio ground relay [13], randomness algorithm
[14], use of current flicker [15], fractal techniques [16], and
signal superposition [17]. Frequency and time domain hybrid
methods such as discrete wavelet transform [18]–[24] have
also been proposed by researchers. Because of the random-
ness of HIFs, expert methods using combinations of multiple
algorithms [25]–[28], multiple tools [29]–[31], Kalman filter
[32], [33] as well as training based methods such as decision
tree [34], artificial neural networks [35]–[38], and neuro-fuzzy
method [39] based on genetic algorithm [40] are proposed.
Another expert method [41] uses neural networks with features
extracted using mathematical morphology for the detection of
HIF.
These prior studies and proposed algorithms have helped re-
veal many hidden properties of the HIF, but they are not ca-
pable of detecting all HIFs. These studies also do not report
the detection delay except for [16] and [27], where the detec-
tion delays are reported as 4 s and approximately 1 min, respec-
tively. This paper proposes a superior method that tracks the
shape of the voltage waveforms to detect an HIF with excellent
reliability and speed, using a tool based on mathematical mor-
phology (MM). No training or learning is required.
MM is a nonlinear time domain signal processing tool that
transforms the shape of signals [42], [43]. The transformation
is based on set theory and integral geometry, and was originally
0885-8950/$31.00 © 2012 IEEE
-
8/18/2019 06311450 (1)
2/9
-
8/18/2019 06311450 (1)
3/9
1228 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 28, NO. 2, MAY 2013
the current waveform, magnitude of the current drawn, and the
frequency content.
Several models have been proposed in the past for the pur-
pose of time domain simulation. A fixed resistance at the point
of fault [2] is the simplest model which was later modified
by including nonlinear impedances [11] to incorporate the
nonlinearity of the fault current. Inclusion of two anti-parallel
dc-sources connected via two diodes [9] modeled the asym-
metric nature of the fault current, as well as the intermediate
arc extinction around current zero. This model was further
modified by adding one [34] or two [20], [33], [37], [39], [50]
variable resistances in series with the dc sources to model
the randomness of the effective impedance and thus the ran-
domness of the resulting fault current. Other models used for
the analysis of HIF are TACS (transient analysis of control
systems) controlled switch to connect and disconnect the fault
[18] or randomly vary the effective HIF resistance [51], and
differential equation based models [21], [22], [38].
This paper uses an HIF model shown in Fig. 3, connected
between phase and ground, being the phasor value of the phase voltage. It is structurally similar to the source-diode-re-
sistance model described in the previous paragraph using cita-
tions; the parameters are tuned to suit the system voltage. The
model is simple, yet captures all the characteristics of HIF cur-
rent. The two dc sources and are connected to two diodes
and , respectively. The dc sources are of unequal magni-
tude, and their valuesrandomly vary around and every 0.1
ms. This arrangement models the asymmetric nature of the fault
current and intermediate arc extinction. The values of and
depend upon the voltage of the system for which the simula-
tion is performed and the amount of asymmetry to be modeled.
When the instantaneous value , current flows towardsground, and reverses when . During the period when
, no current flows. Changing values of and
also add randomness to the amount of asymmetry and the
duration of arc extinction. Two variable resistances and
are also connected in series with the diodes. These resistances
vary independently and randomly every 0.1 ms, and model the
randomly varying arc resistance. To test the robustness of the
proposed algorithm, the range of variation of the resistances is
restricted such that the fault current always lies below 10% of
the full load current of the feeder. Below are the model param-
eters used for HIF simulation with the IEEE 13-node feeder:
%
%
–
–
Fig. 4 shows voltage and current waveforms during a high
impedance fault modeled on the IEEE 13-node test feeder using
the HIF model discussed in the previous paragraph. The current
in Fig. 4(b) is random and asymmetric with unequal positive and
negative peaks. The interruptions around zero crossings repre-
sent temporary arc extinctions. Fig. 5 shows the charac-
teristics of this HIF. The current waveform [Fig. 4(b)] and the
characteristics (Fig. 5) are similar to those obtained from a
staged fault [15], from a lab setup [9], [22] and from other high-
Fig. 3. HIF model.
Fig. 4. Arc voltage and current waveforms during HIF.
impedance arc models [21], [22]. Moreover, the fast Fourier
transform (FFT) analysis of the HIF current waveform yields an
average second harmonic content of 4.8%, and third harmonic
content of 11.8%. These numbers are within the observed range
of similar harmonics during the staged fault reported in [15].
Hence the HIF model in Fig. 3 faithfully represents the nonlin-
earity, randomness and asymmetry in the fault current, as well
as the arc dynamics and intermediate arc extinctions.
C. Selection of Waveform for HIF Detection
The methods proposed for HIF detection either process only
the current waveforms, or both current and voltage waveforms
to obtain signatures that characterize HIFs. However, it should
be kept in mind that the waveform shown in Fig. 4(b) is the
actual fault current, not the current flowing at the substation
source. The current waveform at the substation will be much
less distorted, because the system still draws sinusoidal load cur-
rent. In our case, in order to test our method against the worst
cases, we have limited our HIF current to less than 10% of the
load current. This means the irregularities of the HIF current
may well be masked substantially at the substation. In order to
-
8/18/2019 06311450 (1)
4/9
GAUTAM AND BRAHMA: DETECTION OF HIGH IMPEDANCE FAULT IN POWER DISTRIBUTION SYSTEMS 1229
Fig. 5. V-I characteristics during HIF.
Fig. 6. V-I characteristics during HIF.
examine this, Fig. 6(b) shows the current drawn from the sub-
station during the same HIF. As suspected, this current hardly
shows any traces of the irregularities of the actual HIF current.
The strength of MM based tools lies in the fact that they can
detect and distinguish between very small (seemingly insignif-
icant) changes to a wave-shape. Fig. 4(a) shows the voltagewaveform across theHIF. A zoomed view shows there is a slight
distortion in the waveform. This slight distortion is obviously
also reflected at the substation, as seen in Fig. 6(a). It is possible
that the current waveform at the substation would also contains
a slight disturbance, but the extent of such distortion may vary
for different pre-fault conditions. In contrast, the distortion in
the voltage waveform does not depend on the pre-fault currents,
and is more likely to provide a more consistent signature when
processed by MM. Due to the unique property of MM based
tools to detect very small distortions, and the better likelihood
of the voltage waveforms providing a more consistent signature
when processed by such tools, the voltage waveform is chosen
for processing. The results documented in Sections III and IV
further confirm the validity of this choice.
III. DEVELOPING THE BASIS FOR HIF DETECTION USING MM
A. MM Operations
As mentioned in Section I, MM is composed of two elemen-
tary transformations—dilation and erosion. Based on these two,
several other transformations are defined [52]. Opening and
closing are other commonly used transformations for one-di-mensional signals. These four transformations are defined now.
Equation (1) defines the dilation of a signal by :
(1)
Similarly, (2) defines the erosion of a signal by :
(2)
In these equations, is the signal to be transformed, de-
fined in domain , and is the struc-turing element, defined in domain , and
and are integers such that .
Based on these two elementary transformations, opening and
closing are defined by (3) and (4):
(3)
(4)
It can be observed from (1)–(4) that these transformations
require only addition and comparison. So, they impose a very
low computation burden, which is a great advantage for real
time applications.
The structuring elements are the foundation of all MM trans-
formations, and are used as probes for feature extraction. They
may have different lengths and can be linear, sinusoidal, square,
circular, or other geometrical shapes [52]. The frequency of in-
terest plays a major role in the selection of a structuring element
for a particular application, though the choice is influenced by
other factors such as the type of signal, frequency spectrum, and
the sampling rate. The optimal choice would be such that it cap-
tures the feature of interest while suppressing other features.
Equation (5) defines a Closing Opening Difference Opera-
tion (CODO) [48]. This operation is very effective in detecting
any disturbance in waveforms. The application proposed in this
paper uses the CODO operation to detect and classify HIFs. Alow sampling rate of 3840 Hz (64 samples per cycle) is chosen
so that the computation burden is low, while still preserving the
ability to capture the disturbance signature:
(5)
Gautam and Brahma [53] present a detailed analysis that
leads to some guidelines in selecting an optimal structuring
element to detect disturbances in power system. Based on these
guidelines, samples of voltage waveforms from the secondary
of voltage transformer (VT) are selected to be treated with
CODO. The waveforms are normalized by the peak value of
their rated value prior to the treatment with CODO to make
the application general and applicable at all voltage levels.
-
8/18/2019 06311450 (1)
5/9
1230 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 28, NO. 2, MAY 2013
Fig. 7. Phase A voltage waveform and the CODO output for High ImpedanceFault.
Since the waveforms are normalized, the height (magnitude) of
each element of the structuring element is set at 0.01 [53]. The
guidelines also recommend structuring elements with shorter
length for disturbance detection. So, an iterative process was
adopted to determine an optimal length. Structuring elements
with lengths 2 to 5 were implemented and performance was
observed. It was found that all the structuring elements were
capable of detecting the HIFs, but the shortest one resulted in
the least delay. Therefore, a structuring element of length 2 was
selected for this application.
B. Deriving Unique HIF Signature From the CODO Output Simulations were carried out for high impedance faults and
other disturbances at different locations indicated in Fig. 1. The
legends in the location boxes denote the number of the simula-
tion case with letter prefixes. The prefixes H, C, and L denote
HIF, capacitor switching, and load switching, respectively. For
example, H-2 denotes the second simulation case of HIF. The
voltage waveforms are sampled from the secondary side of VT,
and normalized by the peak value of the rated secondary voltage.
The normalized voltage is then treated with the CODO opera-
tion defined by (5), with the structuring element of length 2 and
height 0.01, selected as described in Section III-A.
Figs. 7–9 show the normalized voltage waveforms and thecorresponding CODO output for three different disturbances.
Since the disturbances in the voltage waveforms are very small,
zoomed parts of the waveforms are inserted to clearly show the
disturbances.
Fig. 7 shows phase A voltage waveform, and the corre-
sponding CODO output for a phase A to Ground HIF simulated
at 6.817 s at location H-4 (Fig. 1). The HIF generates unequal
and non-uniformly distributed series of spikes in the CODO
output over an extended period of time. These spikes are
distinctly larger in magnitude compared to the CODO output
during the healthy condition, which is almost zero.
Fig. 8 shows phase A voltage waveform, and the corre-
sponding CODO output for a three phase capacitor switching
(ON) simulated at 6.8043 s at location C-2 (Fig. 1). The CODO
Fig. 8. Phase A voltage waveform and the CODO output for Capacitor Switching.
Fig. 9. Phase A voltage waveform and the CODO output for Load Switching.
output for the disturbance caused by capacitor switching shows
two continuous large spikes, when the disturbance occurs. It
was observed in other cases of capacitor switching reported in
Section V that capacitor switching gave either a single spike or
a series of consecutive spikes, which is distinctly different than
the non-uniformly distributed spikes generated due to an HIF.Since the effect of capacitor switching on voltage waveforms
is short-lived, the duration of such continuous spikes was never
observed to be more than one eighth of a cycle, as opposed to
the spikes corresponding to an HIF, which continued over time.
Fig. 9 shows phase A voltage waveform, and the corre-
sponding CODO output for a two-phase (A-C) load switching
at 6.8043 s at location L-1 (Fig. 1). The switching is captured
by a single, relatively smaller spike in the CODO output, as the
effect of load switching on the voltage waveforms is minimal.
Figs. 7–9 show that the inception of disturbances are cap-
tured by one or more distinct high magnitude spikes in the
output of the CODO operation, which otherwise lies below
a low threshold during healthy condition. The observation of
the CODO output of the three disturbances also reveal that the
-
8/18/2019 06311450 (1)
6/9
GAUTAM AND BRAHMA: DETECTION OF HIGH IMPEDANCE FAULT IN POWER DISTRIBUTION SYSTEMS 1231
pattern of a series of spikes generated by the HIF is clearly
different from the single or continuous spikes over a short time
span generated by capacitor switching, and load switching. This
distinction in signatures can be exploited to detect and classify
HIF and non-HIF disturbances. The detection and classification
method is described in the following section.
IV. FORMULATION OF METHOD TO DETECT AND
CLASSIFY A HIGH IMPEDANCE FAULT
This section describes a method to detect an HIF, and distin-
guish it from other disturbances discussed in Section III. The
first step in this process is the detection of disturbance. Based
on the analysis of the CODO output for the three different dis-
turbances in Section III, a disturbance is detected if a spike is
present in the output of the CODO operation. A threshold value
is set and the CODO output is tracked for each phase-voltage
to detect the disturbance. For the application proposed in this
paper, the threshold value was set at 115% of the maximum
value of the CODO output during the healthy condition. As
mentioned in Section III, the CODO output is close to zeroduring healthy condition; the 15% margin is simply a buffer.
This threshold value translates to approximately 0.0053 pu for
the examples shown in Figs. 7–9.
The next step is to distinguish an HIF from other distur-
bances. As mentioned in Section III, the comparison of the
CODO output for different disturbances in Figs. 7–9 shows that
an HIF generates a series of non-uniformly distributed spikes
over an extended time-period, whereas a capacitor switching
and a load switching generate either a single spike or multiple
consecutive spikes over a short time period never longer than
one eighth of a cycle. The difference in the signatures is so
clear, there is no need for any learning based pattern recogni-tion method. A rule-based method is therefore adopted. This
method, and an algorithm to implement it are described now.
In addition to the threshold value defined previously in this
section to detect a disturbance, the algorithm requires two other
parameters—wait-time and reset-time —for the clas-
sification of HIF. Wait-time is implemented to avoid the
initial multiple consecutive spikes that are sometimes generated
by capacitor switching. To make sure such spikes are no longer
present in the CODO output, a conservative choice of a quarter
cycle is made for the wait-time. For the chosen sampling rate of
64 sample per cycle, the wait-time translates to 16 samples, and
equals to 4.17 ms in a 60-Hz system.Once the wait time is over, the method prepares to detect an-
other spike in the CODO output. If such a spike is detected, it
is certainly due to an HIF. The detection time for this second
spike, or the detection delay, depends upon the amount of tran-
sients present in the voltage waveform. A slower change in the
effective impedance of HIF implies less transients in voltage
waveform and thus, a larger separation of spikes, and a larger
detection delay. Similarly, a faster change implies more tran-
sients and thus, a smaller separation of spikes, and a smaller
detection delay. Therefore, it is necessary to wait for some time
to allow even the most sparsely separated spike to be detected.
This time is defined as the reset time . If no spike is en-
countered during the reset time, the algorithm determines the
detected disturbance was not an HIF, and resets itself. However,
Fig. 10. Flowchart for the proposed detection and classification method.
if a spike is detected between the end of the wait time and end of
the reset time, the HIF flag is made high. At the end of the reset
time, if the HIF flag is high, a trip or alarm signal is generated,otherwise the algorithm resets and starts afresh. Based on the
simulation studies, a conservative choice of 1 s is made for the
value of the reset time . This means any HIF will be detected
and classified within 1 s.
The algorithm that captures this logic is shown in Fig. 10.
As can be seen from this flowchart, at the very beginning of
the process, a pickup signal from an overcurrent relay is in-
cluded. If the relay picks up, it means there is a conventional
fault, and the whole algorithm is simply suspended by resetting
all flags and timers. Thus, the proposed detection and classifi-
cation method is integrated with a digital overcurrent relay, and
simply suspends operation for all faults detected by the over-
current relay. The rest of the algorithm is self-explanatory, as it
basically implements the detection and classification logic de-
scribed in the previous paragraphs. The threshold value Th is
taken as described previously in this section. DF stands for Dis-
turbance Flag.
V. SIMULATION R ESULTS
A total of 15 HIF cases were simulated at different loca-
tions, on different phases and at different time-instants in the
IEEE 13-node test feeder. The HIF locations were chosen at
nodes as well as at intermediate points. Intermediate points were
chosen in the branches with distributed loads. The locations
were chosen to include 1-phase, 2-phase, and 3-phase branches;
and both unbroken (denoted as type-1) and broken (denoted
-
8/18/2019 06311450 (1)
7/9
1232 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 28, NO. 2, MAY 2013
TABLE ISIMULATION R ESULT FOR IEEE 13-NODE TEST FEEDER
as type-2) conductors were considered. Six cases of capacitor
switching and four cases of load switching were also simulated.The simulation consisted of switching ON and OFF of the ca-
pacitor banks at node 611 and node 675. Capacitor switching
also included switching the banks ON and OFF at different time-
instants (at different points on the voltage waveform). Similarly,
load switching consisted of switching ON and OFF of 2-ph and
3-ph loads at node 692 and node 671 at different time-instants.
The different disturbance inception times were considered since
the transients in the voltage waveforms are dependent on in-
ception times. All these cases are marked out in Fig. 1, with
legend explained in Section III-B. The algorithm described in
Section IV was implemented on the sampled voltage waveforms
from the VT secondary. Table I summarizes the results.The table shows the disturbances simulated along with their
location, type (nature), and inception time. For the HIF cases,
the time of occurrence of the second nonconsecutive spike, and
the corresponding detection delay are also shown. For other dis-
turbances, there is no second spike detected. As seen from the
table, the algorithm successfully detects all cases of HIF while
segregating all other disturbances. The detection delay varies
from 17 ms for the fastest detection to 588 ms for the slowest
one.
The study was then carried out for the 34-node feeder. The
process of selecting disturbances, their locations, and types was
the same as for the 13-node case. Fig. 2 shows all the distur-
bances simulated. Table II summarizes the results for 34-node
feeder, which also shows excellent detection and classification
TABLE IISIMULATION R ESULT FOR IEEE 34-NODE TEST FEEDER
of HIFs. The detection delay in this case varies from less than
10 ms for the fastest detection to 66 ms for the slowest one. As
mentioned in Section II-A, the 34-node feeder is lightly loaded.This means the method is not affected by loading conditions.
To test the reliability and robustness of the method even further,
a study was performed with both test systems operating at half
load. The performance of the method was similar to that ob-
served with full load. This underscores the rationale described
in Section II-C in choosing the voltage waveform for this study,
making the method robust against pre-fault (load) conditions.
These results show that the algorithm is robust, since it success-
fully works on two standard distribution systems with very dif-
ferent properties, simulated with varying load conditions. The
100% success rate of detection and classification for both sys-
tems indicates the proposed method exhibits excellent reliabilityand security.
VI. CONCLUSION
This paper summarizes the use of MM to detect high
impedance faults which are otherwise undetected by the over-
current protection scheme in a power distribution system. An
MM based tool is proposed, implemented and tested to develop
a method that can be integrated as a separate module with a
digital overcurrent relay. The detection method uses voltage
waveforms sampled at the substation. The proposed method is
designed to operate in parallel and assist the existing protection
scheme to detect an HIF. This method is fast, with a detection
delay of 1 s. The method is shown to exhibit reliability, as all
the HIF cases simulated were detected, even for fault currents
-
8/18/2019 06311450 (1)
8/9
GAUTAM AND BRAHMA: DETECTION OF HIGH IMPEDANCE FAULT IN POWER DISTRIBUTION SYSTEMS 1233
less than 5% of the full load feeder current. It also exhibits
security, since there was no false operation for other distur-
bances in the system. The method benefits from the inherent
advantage of low computation burden of all MM based tools,
which is an advantage for real-time applications. The method
is successfully tested on different standardized test feeders,
with different types and locations of disturbances at different
inception times, and for different pre-fault loading.
R EFERENCES
[1] High Impedance Fault Detection Technology, Mar. 1996, Report of PSRC Working Group D15. [Online]. Available: http://www.pes-psrc.org/Reports/High_Impedance_Fault_Detection_Technology.pdf.
[2] R. Lee and M. Bishop, “A comparison of measured high impedancefault data to digital computer modeling results,” IEEE Trans. Power
App. Syst., vol. PAS-104, no. 10, pp. 2754–2758, Oct. 1985.[3] M. Sedighizadeh et al., “Approaches in high impedance fault detec-
tion—A chronological review,” Adv. Elect. Comput. Eng., vol. 10, no.3, pp. 114–128, 2010.
[4] Detection of High Impedance Faults, Power Technologies Inc., EPRI
Report EL-2413, Jun. 1982.[5] High Impedance Fault Detection Using Third Harmonic Current,
Hughes Aircraft Co., EPRI Report EL-2430, Jun. 1982.[6] K.-Y. Lien et al., “Energy variance criterion and threshold tuning
scheme for high impedance fault detection,” IEEE Trans. Power Del.,vol. 14, no. 3, pp. 810–817, Jul. 1999.
[7] Detectionof Arcing Faultson DistributionFeeders,Texas A&M Univ.,EPRI Report EL-2757, Dec. 1982.
[8] M. Aucoin and B. D. Russell, “Detection of distribution highimpedance faults using burst noise signals near 60 Hz,” IEEE Trans.
Power Del., vol. 2, no. 2, pp. 342–348, Apr. 1987.[9] A. Emanuel et al., “High impedance fault arcing on sandy soil in 15
kv distribution feeders: contributions to the evaluation of the low fre-quency spectrum,” IEEE Trans. Power Del., vol. 5, no. 2, pp. 676–686,Apr. 1990.
[10] D. Jeerings and J. Linders, “A practical protective relay for down-con-ductor faults,” IEEE Trans. Power Del., vol. 6,no. 2,pp. 565–574,Apr.1991.
[11] D. Yu and S. Khan, “An adaptive high and low impedance fault detec-tion method,” IEEE Trans. Power Del., vol. 9, no. 4, pp. 1812–1821,Oct. 1994.
[12] J. Carr, “Detection of high impedance faults on multi-grounded pri-mary distribution systems,” IEEE Trans. Power App. Syst., vol. PAS-100, no. 4, pp. 2008–2016, Apr. 1981.
[13] R. Lee and M. Bishop, “Performance testing of the ratio ground relayon a four-wire distribution feeder,” IEEE Trans. Power App. Syst., vol.PAS-102, no. 9, pp. 2943–2949, Sep. 1983.
[14] C. Benner et al., “Improved algorithm for detecting arcing faults usingrandom fault behavior,” Elect. Power Syst. Res., vol. 17, no. 1, pp.49–56, 1989.
[15] A. Sultan et al., “Detecting arcing downed-wires using fault currentflicker and half-cycle asymmetry,” IEEE Trans. Power Del., vol. 9, no.1, pp. 461–470, Jan. 1994.
[16] A. Mamishev et al., “Analysis of high impedance faults using fractaltechniques,” IEEE Trans. Power Syst., vol. 11, no. 1, pp. 435–440, Feb.1996.
[17] I. Zamora et al., “New method for detecting low current faults in elec-trical distribution systems,” IEEE Trans. Power Del., vol. 22, no. 4, pp.2072–2079, Oct. 2007.
[18] D. C. T. Wai andX. Yibin, “A novel technique forhigh impedancefaultidentification,” IEEE Trans. Power Del., vol. 13, no. 3, pp. 738–744,Jul. 1998.
[19] S.-J. Huangand C.-T. Hsieh, “High-impedance fault detection utilizinga morlet wavelet transform approach,” IEEE Trans. Power Del., vol.14, no. 4, pp. 1401–1410, Oct. 1999.
[20] T. Lai et al., “High-impedance fault detection using discrete wavelettransform and frequency range and rms conversion,” IEEE Trans.
Power Del., vol. 20, no. 1, pp. 397–407, Jan. 2005.[21] M. Michalik et al., “High-impedance fault detection in distribution net-
works with use of wavelet-based algorithm,” IEEE Trans. Power Del.,vol. 21, no. 4, pp. 1793–1802, Oct. 2006.
[22] N. Elkalashy et al., “Modeling and experimental verification of highimpedance arcing fault in medium voltage networks,” IEEE Trans. Di-elect. Elect. Insulation, vol. 14, no. 2, pp. 375–383, Apr. 2007.
[23] N. Elkalashy et al., “DWT-based detection and transient power direc-tion-based location of high-impedance faults due to leaning trees inunearthed MV networks,” IEEE Trans. Power Del., vol. 23, no. 1, pp.94–101, Jan. 2008.
[24] M. F. Akorede and J. Katende, “Wavelet transform based algorithm for high- impedance faults detection in distribution feeders,” Eur. J. Sci.
Res., vol. 41, no. 2, pp. 238–248, 2010.[25] C. Kim and B. Russell, “High-impedance fault detection system using
an adaptive element model,” Proc. Inst. Elect. Eng., Gener., Transm., Distrib., vol. 140, no. 2, pp. 153–159, Mar. 1993.
[26] B. Russell and C. Benner, “Arcing fault detection for distributionfeeders: Security assessment in long term field trials,” IEEE Trans.
Power Del., vol. 10, no. 2, pp. 676–683, Apr. 1995.[27] C. Benner and B. Russell, “Practical high-impedance fault detection
on distribution feeders,” IEEE Trans. Ind. Appl., vol. 33, no. 3, pp.635–640, May/Jun. 1997.
[28] T. Cui et al., “Integrated scheme for high impedance fault detection inMV distribution system,” in Proc. IEEE/PES Transmission and Distri-bution Conf. Expo.: Latin America 2008, Aug. 2008, pp. 1–6.
[29] H. Jabr and A. Megahed, “A wavelet-FIRANN technique for high-impedance arcing faults detection in distribution systems,” in Proc. Int.Conf. Power Systems Transients (IPST05), Jun. 2005, pp. 1–6.
[30] M. Yang et al., “Detection of downed conductor in distributionsystem,” in Proc. IEEE Power Engineering Society General Meeting,2005, Jun. 2005, vol. 2, pp. 1107–1114.
[31] A.-R. Sedighi et al., “High impedance fault detection based on wavelettransform and statistical pattern recognition,” IEEE Trans. Power Del.,vol. 20, no. 4, pp. 2414–2421, Oct. 2005.
[32] A. Girgis et al., “Analysis of high-impedance fault generated signalsusing a Kalman filtering approach,” IEEE Trans. Power Del., vol. 5,no. 4, pp. 1714–1724, Oct. 1990.
[33] S. R. Samantaray and P. K. Dash, “High impedance fault detection indistribution feeders using extended Kalman filter and support vector machine,” Eur. Trans. Elect. Power , 2009.
[34] Y. Sheng and S. Rovnyak, “Decision tree-based methodology for highimpedance fault detection,” IEEE Trans. Power Del., vol. 19, no. 2, pp.533–536, Apr. 2004.
[35] M. Eissa et al., “A new protection detection technique for highimpedance fault using neural network,” in Proc. Large Engineering Systems Conf. Power Engineering, 2006 , Jul. 2006, pp. 146–151.
[36] H. K. Zadeh,“ANN-based high impedance fault detectionscheme: De-sign and implementation,” Int. J. Emerg. Elect. Power Syst., vol. 4, no.2, pp. 1–14, 2005.
[37] S. Samantaray et al., “High impedance fault detection in power dis-tribution networks using time-frequency transform and probabilisticneural network,” IET Gener., Transm., Distrib., vol. 2, no. 2, pp.261–270, Mar. 2008.
[38] M. Michalik et al., “New ANN-based algorithms for detecting HIFs inmultigrounded MV networks,” IEEE Trans. Power Del., vol. 23, no. 1, pp. 58–66, Jan. 2008.
[39] A. Etemadi and M. Sanaye-Pasand, “High-impedance fault detectionusing multi-resolution signal decomposition and adaptive neural fuzzyinference system,” IET Gener., Transm., Distrib., vol. 2, no. 1, pp.110–118, Jan. 2008.
[40] M.-R. Haghifam et al., “Development of a fuzzy inference system
based on genetic algorithm for high-impedance fault detection,” Proc. Inst. Elect. Eng., Gener., Transm., Distrib., vol. 153, no. 3, pp.359–367, May 2006.
[41] M. Sarlak and S. Shahrtash, “High impedance fault detection usingcombination of multi-layer perceptron neural networks based on multi-resolution morphological gradient features of current waveform,” IET Gener., Transm., Distrib., vol. 5, no. 5, pp. 588–595, May 2011.
[42] P. Maragos and R. Schafer, “Morphological fi lters—Part I: Their set-theoretic analysis and relations to linear shift-invariant filters,” IEEE Trans. Acoust., Speech, Signal Process., vol. 35, no. 8, pp. 1153–1169,Aug. 1987.
[43] P. Maragos and R. Schafer, “Morphological filters—Part II: Their rela-tions to median, order-statistic, and stack filters,” IEEE Trans. Acoust.,Speech, Signal Process., vol. 35, no. 8, pp. 1170–1184, Aug. 1987.
[44] G. Matheron , Random Sets and Integral Geometry. New York:Wiley, 1975.
[45] J. Serra , Imag e Analys is an d Math ematical M orphology. New York:Academic, 1982.
-
8/18/2019 06311450 (1)
9/9
1234 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 28, NO. 2, MAY 2013
[46] S. Gautam and S. M. Brahma, “Overview of mathematical morphologyin power systems—A tutorial approach,” in Proc. IEEE Power & En-ergy Society General Meeting, 2009 (PES’09), Jul. 2009, pp. 1–7.
[47] S. Gautam andS. M. Brahma, “Properties of mathematicalmorphology based filter for onlinefiltering of power system signals,” in Proc. North American Power Symp. (NAPS), 2009, Oct. 2009, pp. 1–5.
[48] S. Gautam and S. M. Brahma, “Application of mathematical mor- phology based filters to detect a power swing,” in Proc. IEEE Power and Energy Society General Meeting, 2010, Jul. 2010, pp. 1–6.
[49] Distribution Test Feeders. IEEE PES Distribution System AnalysisSubcommittee. [Online]. Available: http://ewh.ieee.org/soc/pes/dsacom/testfeeders/index.html.
[50] E. Sortomme et al., “Microgrid protection using communication-as-sisted digital relays,” IEEE Trans. Power Del., vol. 25, no. 4, pp.2789–2796, Oct. 2010.
[51] S. Nam et al., “A modeling method of a high impedance fault in a dis-tribution system using two series time-varying resistances in EMTP,”in Proc. Power Engineering Society Summer Meeting, 2001, 2001, vol.2, pp. 1175–1180.
[52] Q. Wu et al., Protective Relaying of Power Systems Using Mathemat-ical Morphology, ser. Power Systems. New York: Springer, 2009.
[53] S. Gautam and S. M. Brahma, “Guidelines for selection of an optimalstructuring element for mathematical morphology based tools to detect power system disturbances,” in Proc. IEEE Power & Energy SocietyGeneral Meeting, 2012, Jul. 2012, pp. 1–6.
Suresh Gautam (S’08) received the Bachelor of Engineering degree and the Master of Sciencedegree in electrical engineering from the Institute of Engineering, Tribhuvan University, Nepal, in 2002and 2007, respectively, and the Master of Sciencedegree in electrical engineering from New MexicoState University, Las Cruces, in 2010, where he iscurrently pursuing the Ph.D. degree.
From 2002 to 2004, he was with Kathmandu En-gineering College, Kathmandu, Nepal, as a lecturer.From 2004 to 2007, he was with Nepal Electricity
Authority, a public utility of Nepal, as an electrical engineer. His research inter-ests are power system protection, digital relaying, power system transients, and power quality.
Sukumar M. Brahma (M’04–SM’07) receivedthe Bachelor of Engineering degree from L. D.College of Engineering, Ahmedabad, India, in 1989,the Master of Technology degree from the IndianInstitute of Technology, Bombay, India, in 1997,and the Ph.D. degree from Clemson University,Clemson, SC, in 2003, all in electrical engineering.
From 1990 to 1999, he was a lecturer in the
Electrical Engineering Department with B. V. M.Engineering College, Vallabh Vidyanagar, India.From August 2003 to June 2007, he was an Assistant
Professor at Widener University, Chester, PA. He is presently an AssociateProfessor and Associate Director of the Electric Utility Management Program(EUMP) at New Mexico State University, Las Cruces.
Dr. Brahma is the past Chair of the IEEE Power and Energy Society’s LifeLong Learning Subcommittee, Chair of the Distribution System Analysis Sub-committee, Secretary of the Power and Energy Education Committee, and is onseveral working groups of the Power System Relaying Committee.