International Journal Of Power System Operation and Energy Management (IJPSOEM) Volume-1, Issue-1, 2011

64
FCM and Statistical Based Approach for Classification and Location of Faults in Electrical Distribution
System
Pratul Arvind
1
, Rudra prakash Maheswari
2
1,2
Department of Electrical Engineering, Indian Institute of Technology, Roorkee, India
e-mail id: [email protected]
1
, [email protected]
2
AbstractElectric Power Distribution System is a complex network of
electrical power system. Also, large number of lines on a distribution
system experiences regular faults which lead to high value of current.
Speedy and precise fault location plays a pivotal role in accelerating
system restoration which is a need of modern day. Unlike transmission
system which involves a simple connection, distribution system has a very
complicated structure thereby making it a herculean task to design the
network for computational analysis. In this paper, the authors have
simulated IEEE 13- node distribution system using PSCAD which is an
unbalanced system and current samples are generated at the substation
end. A Fuzzy c-mean (FCM) and statistical based approach has been
used. Samples are transformed as clusters by use of FCM and fed to
Expectation- Maximization (EM) algorithm for classifying and locating
faults in an unbalanced distribution system. Further, it is to be kept in
mind that the combination has not been used for the above purpose as per
the literature available till date.
Keywords-PSCAD, IEEE 13-node feeder, FCM, EM.
I. INTRODUCTION
Electric Power Distribution System is a complicated network of
electrical power system. Analogous to humans circulatory system if
transmission system can be termed as the arteries of human body then
distribution system are the capillaries. Unlike transmission system
which involves an easy connection, Distribution system [1] comprises
of number of radial feeders which has to be highly reliable and
efficient under normal and contingency condition. Transmission
system had been a broad area for researchers due to its simplified
structure, carries major portion of power over long distances and also
considering the impact of the faults that would have on these kinds of
lines. But presently due to the increased urbanization and
industrialization, the amount of power carried by the distribution
grids has also enhanced quite considerably. The large numbers of
lines in a distribution system experience regular faults which lead to
high value of line current. With the availability of inadequate system
information and presence of high impedance faults, identifying and
locating faults in a distribution system pose a major challenge to the
utility operators. Further, in digital protection schemes, for proper
operation of protective relays, correct determination of fault type is a
prerequisite. Speedy and precise fault location plays a significant role
in accelerating system restoration, reducing outage time and
significantly improving system reliability. The methods proposed for
fault location in transmission lines [2] are not easily applicable to
distribution systems. Ratan Das et-al in [3] presents the design and
development of a prototype fault locator which estimates the location
of shunt faults on radial sub transmission and distribution lines based
on the fundamental frequency component of voltages and currents
measured at the line terminal. A review of the classical techniques
and knowledge based approaches can be noticed in [4] and [5] which
recommend a hybrid approach for locating faults. The methods for
locating faults in electrical distribution systems may be broadly
classified into three categories. The first deals with the methods that
detect components of high frequency in travelling waves, the second
includes methods that compute fault impedance from the rms values
of current and voltages measured at the fundamental frequency, and
the last but not the least is based on methods of visual inspection that
consist of patrolling and checking the faulted feeder [6] and [7].
Several methods have been proposed for fault location [8] in power
distribution systems. Most of them estimate the equivalent distance to
the fault based on the impedance estimation as seen from the
substation. The common drawback of the impedance-based methods
is the multiple-estimation problem given by the existence of multiple
points in the power distribution systems that fulfill the equivalent
impedance condition. Consequently, these methods provide precise
but uncertain fault locations. With the introduction of digital signal
processing tools in power system, wavelet transform came into play
for extraction of current features that can be subjected to algorithm
meant for appropriate location of faults but yet an errorless fault
location could not be achieved. Pratul et al [9] used Gabor transform
to collect the features for determining the thresholds for fault location
in an unbalanced distribution system. N-ary tree structure has also
been proposed in [10] for locating faults due to the highly branched
and the non-homogeneity nature of the distribution systems. Also, a
combination of artificial neural network and support vector machine
can be seen in [11] but training in neural network is itself a
cumbersome process. Current features extracted from wavelet multi
resolution approach [12] have not been able to fetch accurate result.
In the present paper, the authors have simulated IEEE 13- node
distribution system using PSCAD which is an unbalanced system and
current samples are generated at the substation end. The current
samples are subjected to FCM to obtain clusters and fed to
expectation maximization algorithm [13]. The paper presents an
alternative solution to the problems associated with interruptions by
means of a statistical modeling of current sample database applied to
determine the fault location in power distribution systems to reduce
the system restoration time.
II. GENERATION OF CURRENT SAMPLES
One of the crucial blocks in locating fault in a distribution system
involves its design on computer interface software on which algorithm
can be verified. The IEEE 13 node radial feeder [14] shown in Figure.
1 is considered as reference for generation of current samples at the
substation end. The purpose behind publishing IEEE 13 node feeder
data is to make available a common set of data that can be used by
program developers and users so that the appropriateness of their
solutions can be verified. Though the feeder is very small yet it
displays some very interesting characteristics such as it is short and
relatively highly loaded for a 4.16 kV feeder, has one substation
voltage regulator consisting of three single-phase units connected in
wye, overhead and underground lines are also present with variety of
phasing. It is further equipped with shunt capacitor banks, in-line
transformer and unbalanced spot and distributed loads. It is considered
as very unbalanced system.
FCM and Statistical Based Approach for Classification and Location of Faults in Electrical Distribution System
International Journal Of Power System Operation and Energy Management (IJPSOEM) Volume-1, Issue-1, 2011
65
Figure 1: IEEE 13 node feeder
Figure 2: Simulation of IEEE 13- node feeder in PSCAD
The feeder shown in Figure 2 is designed and simulated in PSCAD
[15] which is a powerful electromagnetic transient simulation program
most suitable for time domain simulations of the systems. The
graphical interface of the software makes it very easy to build the
circuit and observe the results within a single integrated environment.
For the purpose of modeling certain assumptions were considered such
as the elimination of voltage regulator, purging of cable, and
distributed load being replaced as spot load at the end of the segment.
The Frequency Dependent (Phase) Model is considered since it is
numerically accurate and robust transmission line model available.
Three Phase current samples is obtained at substation end after
extensive simulation [16] by creating all nine types of faults (single
line to ground fault, double line to ground fault, line line) fault)
respectively at various locations i. e between nodes 632 633, 632
671 at different fault resistance value ranging from 0 ohm to 30 ohm at
a step of 10 ohm and at 180
0
fault inception angle over various
locations between the node. The simulation has the duration of run for
1 second; with fault occurring at 0.34165 sec and duration of fault
being 0.1 sec. the peak absolute values of the current samples are
being considered. The rms values of the current samples are obtained
using PSCAD. Line 632 633 is considered as zone 1 and 632 671
is considered as zone 2 respectively. Various locations are taken
between nodes 632 633, 632 671 in step of 10% of total length of
the line in the respective zones because they are the three phase
connections as per the given data. The samples thus obtained are
utilized as the input for the algorithm presented later.
III. APPROACHFOR FAULTLOCATION
An approach to resolve the problem of fault location in
distribution system subjected to different kinds of fault by examining
system behavior is presented. After simulating a distribution system
with different types fault over a range of fault resistance and at
various locations, current waveforms are recorded at the substation
end. These samples are pre-processed in PSCAD to obtain the rms
value. Each recorded event has relevant information that enables data
classification with certain types of classes established in the model.
Detectable groups were taken into account in a preliminary data
analysis. The goal is set to achieve by associating groups to zones in
order to establish correspondence between fault location and data
classification within the groups. Fuzzy c- mean is then applied on
these current samples thus obtained and are subjected to expectation
maximization algorithm for fault classification and fault location of
zones respectively. A detailed algorithm for FCM and EM-algorithm
applied for the above purpose is presented below:
A. Fuzzy c-Means (FCM) Clustering
Fuzzy c-means (FCM) clustering was developed by Dunn [17] in
1974. This was further generalized by Bezdek [18] in 1981 and has
become popular. It is considered as a derivative of k-means
clustering. Clustering data allows the conformation of meaningful
groups in an analytical way, which helps to classify data according to
similarities or affinities. The clustering algorithms are based on the
use of metric differences for the distance estimation. Various types of
clustering methods have been developed. Out of them fuzzy
clustering finds its vital representation in the field of data mining,
artificial intelligence, numerical taxonomy, pattern recognition,
image analysis, image processing, and medicine,. It is widely used
because of fuzzy membership, since fuzzy sets could allow
membership functions to all clusters in a data set so that it is very
suitable for cluster analysis. Fuzzy C-means algorithm is based on the
FCM and Statistical Based Approach for Classification and Location of Faults in Electrical Distribution System
International Journal Of Power System Operation and Energy Management (IJPSOEM) Volume-1, Issue-1, 2011
66
minimization of a criterion function. FCM clustering algorithm [19]
is applied because of good performance and less execution time to
obtain clustered data. In the proposed work, authors have used
Haojun Sun et al. [20] algorithm to fix the number of clusters.
Suppose a matrix of n data elements (fault signal), each of
size ( 3) s s = is represented as
1 2
( , ,....., ). =
n
X x x x FCM
establishes the clustering by iteratively minimizing the objective
function as given in Eq. (1)
Objective function:
2
1 1
( , ) ( , )
c n
m
m ij j i
i j
O U C U D x C
= =
=

(1)
Constraint:
1
1;
c
ij
i
U j
=
=

(2)
Where,
ij
U is membership of the
th
j data in the
th
i cluster
i
C
,
m
stands for the fuzziness of the system ( 2) m = and D represents the
distance between the cluster center and data point.
B. FCM Algorithm
Flow chart of FCM algorithm is shown in Figure 3. The
implementation steps are given below:
Input: fault signal data; Output: Clustered data;
- Initialize the cluster centers
i
C .
- Calculate the distance D between the cluster center and
data point by using Eq. (3)
2
2
( , )
j i j i
D x C x C = (3)
- Calculate the membership values by using Eq. (4)
1
2 ( 1)
1
( , )
( , )
m
c
j i
ij
k j k
D x C
U
D x C

=
| |
| |
|
= |
|
|
\ .
\ .

(4)
- Update the cluster centers using Eq. (5)
1
1
n
m
ij j
j
i n
m
ij
j
U x
C
U
=
=
=

(5)
- The iterative process starts:
1. Update the membership values
ij
U by using Eq.(4)
2 Update the cluster centers
i
C by using Eq. (5).
3 Update the distance D using Eq. (3).
4 If ; ( 0.001)
new old
C C > = then go to step1.
5 Else stop.
- Assign each fault signal to a specific cluster for which the
membership is maximal.
IV. EXPECTATION MAXIMIZATION ALGORITHM
From information of the groups obtained using the FCM
algorithm (discussed in section III), initial values for the centers are
estimated. The initial value of covariance matrix is taken as the
identity matrix and the mixture coefficients are then calculated with
the proportion of data in each group, in relation to the sample. Once
initial parameters are obtained, the estimation of the mixture model
parameters is initiated by the Expectation - Maximization algorithm
[21], which is an iterative procedure until the desired convergence is
achieved. EM is an iterative approach to maximum likelihood
estimation. Each Iteration of an EM algorithm consists of two steps:
an Estimation (E) step and a Maximization (M) step. The M step
involves the maximization of a likelihood function that is redefined in
each iteration by the E step. The results are the final values of
parameters (mean vector), V (covariance matrix) and p (weight/
coefficient of mixture) of each group. The steps of the Expectation
Maximization algorithm are as follows:
1. Determine the number of components of the mixture by
using the fuzzy cluster-mean algorithm.
2. Determine initial values of parameters of each
component

( )

0 (0) (0)
( , , ) V p .
3. Calculate the posterior probability for each observation
(Expectation-steep) as shown in the following equations:

( ; , )
( )
i
i j i
ij
j
p x V
f x

=

(6)

1
( ) ( ; , )
G
j g j g g
g
f x p x V
=
=

(7)
Where ij

represents the posterior probability of

p of each component (maximization-step)

1
1
n
ij
i
j
p
n

=
=

(8)

1
1
n
ij
j
i
j
i
x
n
p

=
=


(9)

1
( )( )
1
T
n
ij
j i j i
i
j
i
x x
V
n
p

=

=


(10)
5. Repeat steps 3 and 4, until desired convergence is obtained.
Subsequently, the organization of groups in classes associated to
faults is based in the probability of appearance in each group as given
by the mixture model in the following equation:

1
( ) ( ; , )
G
FM g g g g
g
f x p x V
=
=

(11)
FCM and Statistical Based Approach for Classification and Location of Faults in Electrical Distribution System
International Journal Of Power System Operation and Energy Management (IJPSOEM) Volume-1, Issue-1, 2011
67
Where

( )
FM
f x corresponds to mixture model of sample ( ) x which
corresponds to random sample of n observations of dimension d .
The fault location has been done using this expectation
maximization algorithm on the clusters obtained after using FCM. It
is to be mentioned that the result obtained are shown for
classification, and zone identification
Figure 3: Flowchart for FCM algorithm
V. RESULTS AND DISCUSSION
With the statistical model presented in the previous section, fault
location is done as per the response available. Current waveforms
recorded at the substation after extensive simulation in PSCAD acts as
the input. The root mean square (rms) values of the current samples
have been used. The approach is aimed to obtain a low economical
cost due to constraints in most of the distribution utilities. For
obtaining the result, visual discrimination of the zones have been done
in preliminary data analysis. A total of 720 current samples were
collected from an IEEE 13-node feeder, which as per the IEEE is a
very unbalanced system. The current samples were taken for two
zones over a resistance ranging from (0 10) at ten different
locations. It is to be kept in mind that owing to its nature this feeder is
neglected by the researchers. The authors are successful in getting the
results. Tables represent the results shown for one category for all
types of faults such as single line to ground, line to line and double
line to ground fault respectively which are very promising.
The result of the proposed algorithm is presented here. Table 1
furnishes information about fault classification after being identified
with 0 resistance value. 10 sample each from phase a, b and c from
zone 1 and zone 2 respectively are mixed together. This mixture of
samples is fed given as input to FCM to obtain cluster centers. Using
these cluster centre EM algorithm is applied to yield 100%
classification result.
TABLE I. FAULT CLASSIFICATION
Fault Type Phase
(A)
% Phase
(B)
% Phase
(C)
% Sub
-total
%
Single -line to
ground
20/20 100 20/20 100 20/20 100 100
Phase
(AB)
% Phase
(BC)
% Phase
(AC)
%
Line to line 20/20 100 20/20 100 20/20 100 100
Double line to
ground
20/20 100 20/20 100 20/20 100 100
Total 180/180 100
TABLE II. SAMPLES IDENTIFIED UNDER DIFFERENT FAULT RESISTANCE
Fault Type Resistances
R1 % R2 % R3 % Sub -
total %
Single -line
to ground
25/30 83.33 30/30 100 26/30 86.67 90.00
Line to line 27/30 90.00 20/30 66.67 26/30 86.67 81.11
Double line
to ground
25/30 83.33 26/30 86.67 15/30 50.00 73.33
Total 220/270 81.48
TABLE III. ZONE IDENTIFICATION
Fault Type Zones
Z1 % Z2 % Sub -total %
Single -line to
ground
10/10 100 10/10 100 100
Line to line 10/10 100 10/10 100 100
Double line to
ground
08/10 80 07/10 70 75
Total 55/60 91.67
Table 2 give results of the samples obtained from one-phase over
different resistances which are mixed and subjected to above
procedure. Output yields 81.48% of the samples are recognized over
the correct resistances. Table 3 gives information about different
zones. These zones are the different nodes where the current samples
have been obtained. Results are presented for one type of fault in all
the different category of fault. Single- line to ground fault, line to line
fault are exactly located where the fault has occurred. The total zone
identification is 91.67%.
VI. CONCLUSION
Authors have been successful in locating faults in a standard
IEEE 13- node distribution system using PSCAD. It is obvious from
the literature that the feeder is an unbalanced and therefore there has
been problem in locating faults since there exists large variation in
the current magnitude. A valuable methodology has been presented.
The approach is based on statistical modeling of the samples obtained
after been clustered by the use of FCM. Results obtained after the
clusters are administered to expectation maximization algorithm for
the given feeder is promising It should be kept in mind that the
proposed algorithm has not been applied to current samples of
distribution system as per the literature available till date. Also, IEEE
FCM and Statistical Based Approach for Classification and Location of Faults in Electrical Distribution System
International Journal Of Power System Operation and Energy Management (IJPSOEM) Volume-1, Issue-1, 2011
68
13- node has not been considered extensively as the reference system
for collecting samples in order to locate faults.
REFERENCES
[1] IEEE guide for Protective Relay Applications to Distribution Line, IEEE Std.
C37.230, 2007.
[2] Yuan Liao, Fault location using unsynchronized voltage measurements during
fault, Electric Power Components and Systems,vol. 12, pp. 12831293, 2006.
[3] R. Das, M. Sachdev, T. Sidhu, A fault locator for radial sub transmission and
distribution lines, Proceeding of IEEE Power Engineering Society Summer
Meeting, Seattle,vol. 1, pp. 443- 448, 2000.
[4] J. Mora, J. Melndez, et al, An overview to fault location methods in distribution
system based on single end measures of voltage and current, Proceeding of
ICREPQ04 International Conference on Renewable Energy and Power Quality,
Barcelona, April 2004.
[5] M. Mirzaei, M.Z. A Ab Kadir, E. Moazami, H. Hizam, A Review of Fault
Location Methods for Distribution Power System, Australian Journal of Basic and
Applied Sciences, vol. 3, pp. 2670 2676, 2009.
[6] T. Short, Electrical Power Distribution Handbook, CRC Press, 2003.
[7] J. Zhu, D. Lubkeman, A. Girgis, Automated Fault location and diagnosis on
electric power distribution feeders, IEEE Transactions on Industrial Applications,
vol. 2, no. 2, pp. 801 809, 2002.
[8] J. Mora - Florez, et a, Comparism of impedance based fault location methods for
power distribution system, Electric Power System Research, vol. 78, no. 4, pp.
657 666, 2008.
[9] Pratul Arvind, Rudra Prakash Maheshwari, A Gabor Filter Based Approach for
Locating faults in Distribution System, Proceedings of IEEE International
Conference on Control, Robotics and Cybernetics, vol. 2, pp. 309-313, March
2011.
[10] S. Herraiz, J. Melndez, G. Ribugent,et al, Application for Fault Location in
Electrical Power Distribution Systems, Proceeding of 9
th
International Conference
on Electrical Power Quality and Utilisation, 9 -11 October 2007.
[11] D. Thukaram, et al, Artificial neural network and support vector machine
approach for locating faults in radial distribution systems, IEEE Transaction on
Power Delivery, vol. 2,no. 20, pp. 710720, April 2005.
[12] U. D. Dwivedi, S. N. Singh, S. C. Srivastava, A Wavelet Based Approach for
Classification and Location of Faults in Distribution Systems, Proceeding of IEEE
INDICOM, Kanpur, vol.2,pp.488-493, Dec 2008.
[13] J. Mora - Florez, et a, k mean algorithm for locating faults in power system,
Electric Power System Research, vol. 79, no. 4, pp. 714 721, 2009.
[14] IEEE Distribution Planning Working Group Report, Radial distribution test
feeders, IEEE Transaction on Power System, Vol. 6, No. 3, pp. 975 985, August
1991.
[15] PSCAD, User Guide, Manitoba Research Centre, Canada.
[16] Pratul Arvind, Rudra Prakash Maheshwari, Simulation of IEEE 13-node feeder
and Its wavelet decomposition, Proceedings of ICOPS10 in International Journal
of Emerging Technologies and Applications in Engineering Technology and
Sciences (IJ-ETA-ETS), pp. 423 430, December, 2010.
[17] Dunn, J. C., A Fuzzy Relative of the ISODATA Process and Its Use in Detecting
Compact Well-Separated Clusters, Taylor & Francis Journal of. Cybernetics and
Systems, 3 (3) (1974) 3257.
[18] J. C. Bezdek, Pattern Recognition with Fuzzy Objective Function Algorithms,
Plenum Press, 1981, New York.
[19] J. Hair, R. Anderson, R. Tatham, W. Black, Multivariable Data Analysis, Prentice
Hall, Madrid, 1999.
[20] Haojun Sun, S. Wang and Q. Jiang, FCM-Based Selection Algorithm for
Determining the Number of Clusters, Elsevier J. Pattern Recognition, vol. 37, pp.
2027-2037, 2004
[21] M. Jordan, R. Jacobs, Hierarchical mixture of experts and the EM algorithm,
IEEE Transactions on Neural Computation, vol.6, pp. 181-214, 1994