A Wavelet- Based Fault Localization in

Transmission Network
Sushma Verma, Pratyay Konar, and Dr. Paramita Chattopadhyay.

transform approaches. Due to this they have received great

Abstract-- The proposed technique consists of a pre-processing attention in the power community in the last few years. The
unit based on Continuous Wavelet Transform (CWT) in potential benefits of applying Wavelet Transforms in
combination with an Artificial Neural Networks (ANN). CWT combination with soft computing techniques for improving the
acts as an extractor of distinctive features in the transient current
signals at the sending and receiving end of the transmission line. performance of protective relays have already been recognized
This information is then fed to the ANN for detecting the fault by various researchers. In the mainstream literature, wavelets
type and location of fault. The results presented clearly indicate were first applied to power system in 1994 by Robertson and
that the present technique is very fast, computationally efficient Ribeiro as reported in [5]. From this year the number of
and intelligent enough to accurately identify three different types publications in this area has increased exponentially. There
of fault (LG & LLG) and their locations.
are various approaches for analysis of power system faults
based on expert system [1],[3],[4] fuzzy logic [6],
Index Terms-- Artificial Neural Network (ANN), Alternate
Transient Program(ATP), Continuous Wavelet Transform
optimization [2], pattern recognition approach [7], artificial
(CWT), Fault location, Wavelet Transform, Transmission line . neural network (ANN) [8], wavelet techniques [9], etc. Each
of them has some difficulties in achieving speed, selectivity
I. INTRODUCTION and accuracy. Discrete wavelet transform in combination with
HE electrical power system network is spread widely ANN has also been implemented for this purpose. However

T throughout the geographical regions. The demand of

power is growing with the development of the country.
there is no report of using Continuous Wavelet Transform
(CWT) in classification of power system transients. CWT has
The growing demand leads to the interconnection of the power great potential in analyzing system transients.
networks. As the number of interconnections is increasing
II. PROPOSED SCHEME
everyday the existing power system has become increasingly
complex. There are various equipments in the power system Keeping the above viewpoints in mind the present work has
network at both the ends. The exposure of overhead line been directed towards the development of CWT-ANN based
experiences a permanent or short duration fault. There are fast, intelligent and reliable algorithm to discriminate and
various causes like capacitor switching, lightning, high identify different power system transients. The work has been
impedance faults etc. that leads to transient disturbances in mainly focused on identification of different types of
transmission lines. These transients pose a great challenge to transmission line faults along with their locations. The
the power system engineers. The increasing complexity of the schematic representation of the proposed work is shown in the
modern power system network requires a fast fault detection fig. 1.
technique by making the transient an important phenomenon.
The fault current has the fundamental, higher order harmonics
and time-domain information. Analysis of these higher order
harmonics and their time information plays a very important
role for the detection of faults. The Wavelet Transforms are
best suited for the analysis of transient waveforms than other

S. Verma is with Techno India College of Technology ,Rajarhat District

Kolkata, India (e-mail: [email protected]).

P.Konar is with the Department of Electrical Engineering, B.E.S.U,Shibpur

University, Howrah, India (e-mail: [email protected])

P.Chattopadhyay is with the Department of Electrical Engineering,B.E.S.U,

Shibpur, Howrah, India (e-mail: [email protected]

978-1-4673-0136-7/11/$26.00 ©2011 IEEE

Fig. 1 Block diagram of the proposed scheme
 III. SIMULATION OF TRANSMISSION LINE MODEL signal into shifted and scaled versions of the original (or
In this study a power system network consisting of two mother) wavelet. So, sharp changes might be better analyzed
three phase voltage sources are used. The length of the with an irregular wavelet than with a smooth sinusoid.
transmission line is 80 km. Therefore, to summarize the procedure of wavelet analysis
may run like this – a wavelet prototype function, called an
analyzing wavelet or mother wavelet is adopted and then
temporal analysis is performed with a contracted, high-
frequency version of the prototype wavelet, while frequency
analysis is performed with a dilated, low frequency version of
Fig. 2 Single line diagram for system under study the same wavelet. Thus the original signal or function can be
represented in terms of a wavelet expansion (using coefficients
The power transmission system was modeled with EMTP in a linear combination of the wavelet functions). Data
to test the performance of the proposed scheme. The single operations can be performed using just the corresponding
line diagram for the system under study is presented in Fig. 2 wavelet coefficients.
along with the EMTP/ATP model in Fig. 3.

Fig. 3 EMTP Model for 80 Km line with single line to ground fault

Extensive simulations were carried out, with different fault

locations , for single line to ground fault (LG) and double line Fig. 5 Phase ‘a’ current for single line to ground fault as recorded on
to ground fault (LLG) for different fault resistances (10Ω, 50Ω receiving end side.
and100Ω). Three phase transient current signals from both
sending and receiving ends were captured using ATP/ EMTP The continuous wavelet transform (CWT) is defined as the
simulation environment. The current wave forms were sum over all time of the signal multiplied by scaled, shifted
generated at a sampling frequency of 2 KHz as shown in Figs. versions of the wavelet function Ψ.
4 and 5.
Lψ f ( s,τ ) = ∫ f (t )Ψs*,τ (t ) dt (1)

f (t) is decomposed into a set of basis function ψ (t), called

wavelets generated from a single basic wavelet ψ(t), the so
called mother wavelet, by scaling and translation

1  t −τ  (2)
Ψs ,τ (t ) = Ψ 
s  s 

s is a scale factor, τ is the translation factor and the factor

s is for energy normalization across the different scale.

Fig. 4 Phase ‘a’ current for single line to ground fault as recorded on sending Scaling a wavelet simply means stretching or compressing it.
end side. In this work CWT has been used as a signal processing
tool, because CWT has a great potential in analyzing system
A wavelet is a waveform of effectively limited duration that transients. The raw data of the transient current signals were
has an average value of zero. Comparing wavelets with sine imported in the MATLAB workspace and CWT coefficients
waves, which are the basis of Fourier analysis, it can be were computed for ‘coif1’ mother wavelet using wavelet tool
appreciated that sinusoids are smooth and predictable; box commands. Typical CWT coefficient plots of the transient
wavelets tend to be irregular and asymmetric. Fourier analysis signals are shown in Fig 6 and Fig 7. For all simulation step
consists of breaking up a signal into sine waves of various length of moving window was set to 1/8th cycle (50 Hz
frequencies. Similarly, wavelet analysis is the breaking up of a
supply). Finally the RMS vale of the CWT coefficient was to the input layer through the hidden layer in the network to
computed. The R.M.S values of CWT coefficients thus obtain the final desired outputs. During the forward pass all
obtained were used as attributes to train the Artificial Neural the weights of the networks are fixed. During the backward
Network (ANN). The ANN can easily identify the types of pass, on the other hand, all the weights are adjusted in
faults and their locations. accordance with an error-correction rule.

V. RESULTS AND ANALYSIS

The architecture of the developed wavelet based ANN
models have been presented in Fig 8 (a) & (b).

Fig. 8(a) Neural Network Architecture fault classification

Fig. 6 CWT Phase ‘a’ for Sending end current for single line to ground fault.

Fig. 8(b) Neural Network Architecture fault localization

They have been used for classifying types of fault and to

determine fault locations of the three phase transmission
networks. The developed ANNs have six input neurons
namely RMS values of CWT coefficients of 3-ph sending and
receiving end currents. The ANN models used are shown in
fig. 8(a) and 8(b). The ANN model has three output neurons
for fault classification and one output neuron for fault
Fig. 7 CWT Phase ‘a’ for Receiving end current for single line to ground
fault. localization. The activation functions at the hidden layers and
output layer in the network have been tan-sigmoid and log-
IV. FAULT TYPE IDENTIFICATION AND FAULT LOCATION sigmoid respectively.
DETECTION USING ANN Total 180 feature samples were simulated for different
Post Processing and Diagnosis of faults was done by using switching, single line to ground (LG) and double line to
Artificial Neural Network (ANN) as a classifier. In this work, ground fault (LLG) fault conditions at different locations.
a two layer feed forward neural network has been trained by Among them 120 samples were used for classification purpose
back propagation algorithm. and remaining 60 samples for testing the classifier’s
A feed-forward back-propagation neural network has three generalization ability. From the test results furnished in Table I
components: an input layer, one or more hidden layers, and an and Table II implies that wavelet based techniques correctly
output layer. Each layer consists of one or more neurons called classify the types of faults and their locations also.
nodes. In the calculation process of problem solving, all input TABLE I
nodes are collected at each hidden node after being multiplied FAULT CLASSIFICATION ACCURACY
by weights. Later, a bias is attached to this sum, transformed
through a nonlinearity function, and transferred to the next Type of Fault Training Testing
Accuracy Accuracy
layer. There are several functions such as hyperbolic tangent,
sigmoid and linear functions that can be used as transfer
L-G Fault in phase ‘a’ at 55 km
function. The same procedure can be followed in the next (50 Ω)
99.6% 98.1%
layer to provide the network output results. As the forward L-L-G Fault in phase ‘a’ and ‘b’ at
99.3% 99.4 %
processing arrives at the output layer, the overall error 45 km (100 Ω)
between the network output and the actual observation is L-L-G Fault in phase ‘c’ and ‘b’ at
98.3% 99.2%
calculated. The error at the output layer propagates backward 35 km (50 Ω)
 TABLE II Vector Machines (SVM) or rough set theory can be applied
FAULT LOCALIZATION ACCURACY
for improved classifications or prediction and to overcome the
limitations of ANN may be exploited.
Training Testing Type
of Fault
Desired Network Desired Network VII. ACKNOWLEDGMENT
Output Output Output Output The authors are thankful to Mr. Vijaykumar Garlapati for his
efforts towards this project.
10 10.81 10 9.91 L-G Fault
(phase ‘a’)
10 10.11 10 10.09 VIII. REFERENCES
20 19.99 20 18.31 L-L-G Fault [1] C. Fukui, J. Kawakami, “An expert system for fault section estimation
(phase ‘ab’) using information from protective relays and circuit breakers”, IEEE
20 20.10 20 20.41 Transaction on Power Delivery, vol. 1, pp. 83–90, 12 Oct 1986.
[2] C.S. Chang, L. Tian, F.S. Wen, “A new approach to fault section
30 30.02 30 30.12 L-L-G Fault estimation in power systems using ant system”, Electrical Power
(phase ‘bc’) System Research, vol. 41, pp 63–70, 15 Feb 1999.
30 29.84 30 28.97
[3] Park Young Moon, Kim Gwang-Won, “A logic based expert system for
40 40.09 40 40.59 fault diagnosis of power system,” IEEE Trans Power Syst, Vol 2, pp
L-L-G Fault
363–369, Feb. 1997.
(phase ‘ca’)
40 39.87 40 38.89 [4] Ernesto Vazquez M, Oscar L, Chacon M, Hector J, Altuve F, “An online
expert system for fault section diagnosis in power system”, IEEE
50 50.22 50 49.91 L-G Fault Transactions on Power Systems, vol. 12, no. 1, pp.357 - 362 , 1997.
(phase ‘c’) [5] RM de Castro Fernández, Horacio Nelson Diaz Rojas, “An overview of
50 49.29 50 50.10
wavelet transforms application in power systems”, 14th PSCC, Sevilla,
60 60.03 60 59.76 24-28 June 2002.
L-G Fault [6] C.K. Jung, K.H. Kim, J.B. Lee, B. Klockl, “Wavelet and neuro-fuzzy
(phase ‘b’) based fault location for combined transmission systems”, International
60 59.89 60 60.52
Journal of Electrical Power & Energy Systems, vol 29, pp 445–454.
65 64.95 65 64.71 L-L-G Fault July 2007,
(phase ‘ca’) [7] P.K. Dash, S.R. Samantaray, “A novel distance protection scheme using
65 64.88 65 65.39 time–frequency analysis and pattern recognition approach”,
International Journal of Electrical Power & Energy Systems, Vol.
70 69.91 70 70.55 L-L-G Fault 29(2), pp 129-137, Feb. 2007.
(phase ‘ca’) [8] P.S Bhowmik, P.K Purakit and K.Bhattacharya , “A novel wavelet
70 70.19 70 70.80
transform aided neural network based transmission line fault analysis
75 74.94 75 75.06 L-L-G Fault method ”, International Journal of Electrical Power & Energy Systems,
(phase ‘ca’) vol. 31, pp 213-219, June 2009
75 74.93 75 74.88 [9] Michalik M, Rebizant W, Lukowicz M, Lee S-J, Kang S-H, “Wavelet
transform approach to high impedance fault detection in MV networks”,
80 79.91 80 79.91 L-G Fault Proceedings of the 2005 IEEE power tech conference, St. Petersburg,
(phase ‘b’) Russia, pp. 1-7, 27–30 June, 2005.
80 79.91 80 79.89

IX. BIOGRAPHIES
VI. CONCLUSIONS
In the present era of open access and deregulated electricity Sushma Verma born, on Dec 17 , 1976 is currently
market, it is very important to restore the system after fault working as an Assistant Professor in Techno India
college of Technology in Electrical Engineering
clearance. Accurate identification of fault and its location is
Department, Rajarhat, Kolkata. She graduated from
thus of prime concern, where the transient signals can play a Government Engineering College, Bhuj, Gujarat., India
very vital role. This work is a preliminary report of a research and completed Masters Degree from Guru Nanak Dev
project with a long term goal to utilize the power system Engineering College, Punjab from Punjab Technical
University.
transient signals for fast and accurate discrimination of After her graduation she worked as a Trainee
different switching conditions. The present study demonstrates Engineer in Standard Electricals Limited Company,
the effectiveness of transient current signals to identify the Jalandhar, Punjab.She served as an expatriate lecturer in Royal Bhutan
Institute of Engineering and Technology, Bhutan, for two years. Her special
fault type and fault locations of three phase system. The fields of interest are Power Sytem Transients and Power System Protection.
properties of the proposed scheme are: She is currently working under the esteemed guidance of Dr. Paramita
• The Continuous wavelet transform (CWT) provides an Chattopadhyay.
efficient way to extract important fault features. Pratyay Konar received B.Tech. in Electrical
• Neural network provides an intelligent method and a soft Engineering from Asansol Engineering College under
criterion for feature comparison. West Bengal University of Technology, India in 2007
and M.E. in Electrical Engineering with Electrical
However there is ample scope of work. Variation of pre- Machine specialization from Bengal Engineering and
fault load and system conditions and fault resistances, best Science University, Shibpur, India in 2009. He is
feature selection of wavelet coefficients, impact of mother- currently working to pursue Ph.D. in Bengal
wavelet, variation of scale may be explored in future. Apart Engineering College and Science University, Shibpur,
India. His research interests include image processing
from the ANN other soft-computing techniques like Support
and application of advanced signal processing and applied soft computing
technique to condition monitoring of electrical machines.

Paramita Chattopadhyay received the B.E., M.E.,

and Ph.D. in Electrical Engineering from Bengal
Engineering College and Science University,
Shibpur, India in 1993, 1996, and 2002, respectively.
She is an Assistant Professor at Bengal Engineering
and Science University, Shibpur, India in Electrical
Engineering Department where she specializes in the
field of condition monitoring, power system
protection and power system transient analysis. Her
research interests are applications of advanced signal
processing and soft computing technique in the area of condition monitoring
of electrical machines, power systems and nano material applications in
power sectors.