INTERNATIONAL

International JOURNAL
Journal of Electrical OF ELECTRICAL
Engineering and Technology ENGINEERING
(IJEET), ISSN 0976&–
6545(Print), ISSN 0976 – 6553(Online) Volume 3,
TECHNOLOGY (IJEET)Issue 2, July- September (2012), © IAEME

ISSN 0976 – 6545(Print)

ISSN 0976 – 6553(Online)
Volume 3, Issue 2, July – September (2012), pp. 94-102
IJEET
© IAEME: www.iaeme.com/ijeet.html
Journal Impact Factor (2012): 3.2031 (Calculated by GISI) ©IAEME
www.jifactor.com

TRANSMISSION LINE FAULT CLASSIFICATION BASED ON

WAVELET ENTROPY AND NEURAL NETWORK

Soumyadip Jana¹, Sudipta Nath², Aritra Dasgupta³

¹Department of Electrical Engineering, Techno India, EM 4/1 Salt Lake, Sec-V, Kolkata-
700091, India
²Department of Electrical Engineering, Netaji Subhash Engineering College, Techno City,
Garia, Kolkata-700152, India
³Department of Electronics & Instrumentation Engineering, Netaji Subhash Engineering
College, Techno City, Garia, Kolkata-700152, India
E-mail: soumyadipjana@gmail.com, nath_sudipta@yahoo.com, aritradgupta@yahoo.co.in

ABSTRACT
Classification of faults in power system has been a major challenge for power system
engineers. In this paper, a technique for fault classification in transmission lines has been
developed. EMTP software has been used to simulate the transmission line model. The
simulation is followed by analysis of the voltage waveforms for classification of faults system
using wavelet transform and artificial neural network in the MATLAB environment. The
entropy values of the voltage signal obtained from wavelet transform have been fed into the
neural network for automatic fault type classification. It has been found that only three level of
decomposition of the voltage signal is sufficient for classification of faults. The proposed
scheme is tested under different types of faults, such as single line to ground faults, line-to-line
faults, double line to ground faults and three phase symmetrical faults for different varying
fault locations and fault resistances.
Index terms: Entropy, Faults, Probabilistic neural network, Wavelet transform

1. INTRODUCTION

Transmission and distribution networks are the important components of power system
network. Electric power systems are constantly affected with faults which disturb the system’s
reliability, security and delivered energy quality. For the economic operation and stability of

94
International Journal of Electrical Engineering and Technology (IJEET), ISSN 0976 –
6545(Print), ISSN 0976 – 6553(Online) Volume 3, Issue 2, July- September (2012), © IAEME

the power network, it is of vital importance to detect, classify and clear the transmission line
faults as quickly as possible. The importance of protection of power system network plays a
great role in the successful operation of power system. In power system protection schemes
classification of fault types in a radial and non radial system are the important issues. This can
be done by detecting, localizing and classifying different fault types.
Safty and Zonkoly [1] presented a paper where EMTDC/PSCAD software has been used to
detect fault types and wavelet entropy principle is implemented to analyze the current signals.
Routray et al. [2] proposed a real time wavelet-fuzzy combined approach for digital relaying.
Jung et al. [3] used a MATLAB based simulation tool to calculate the short circuit faults in
power system transmission lines. The discrete wavelet transform has been used along with the
fuzzy logic system to detect the fault classification of a transmission line. Kim et al. [4]
presented a fault location algorithm using the Neuro-fuzzy systems in a combined
transmission lines along with underground power cables. A fault location method employing
wavelet fuzzy neural network has been proposed by Chunju et al. [5]. A new one-end fault
location method for overhead transmission lines embedded in a general n-bus interconnected
system has been discussed by Eisa et al. [6]. Jain et al. [7] proposed the problems encountered
by conventional distance relays when protecting double-circuit transmission lines. Salim et al.
[8] presented a paper in which fault location on transmission lines has been detected by
impedance calculation method. A novel approach for robust fault detection and identification
has been proposed by Zan et al. [9]. Borghetti et al. [10] presented a paper based on a
continuous wavelet transform for the analysis of voltage transients due to line faults, and
discussed its application to fault location in power distribution systems. Magnago et al. [11]
identified the fault location on the parallel transmission lines using wavelet. Saleh et al. [12]
proposed a differential equation to obtain the transient and steady state condition analysis.
Fault location in transmission line is presented by Ekici et al.[13]. Chanda et al. [14] presented
a multi resolution analysis (MRA) for locating the point of strike of lightning in transmission
lines.
In this paper, half cycle of the pre fault and half cycle of the post fault voltage signal has been
considered and decomposed by wavelet transform up to level 3. It has been found that the
entropy values of detail coefficients of level 1 and level 3 only are sufficient for the detection
and classification of faults in both a radial and non radial power system network. The entropy
values have been fed into the neural network for automatic fault classification. The proposed
protection scheme is tested under different fault types, varying fault location and varying fault
resistances.

2. POWER SYSTEM MODEL

Two types of power system network have been considered here. In the first case a single
transmission line has been considered and in the second case a six bus eight line non radial
power system network has been considered. The simulation has been done in Electro
Magnetic Transient Program. This program allows us to simulate the faults of any types at any
location along the transmission line. In the first case a 400KV, 150Km long, three phase
transmission line as shown in Fig. 1 has been considered. The voltage waveforms are
monitored from the sending end of the transmission line. The circuit consists of an ac voltage
source of 400KV, 50Hz which is supplying power to the network throughout the entire period.
15 LCC 3 phase blocks of 10km are taken for simulating 150km long overhead transmission

95
International Journal of Electrical Engineering and Technology (IJEET), ISSN 0976 –
6545(Print), ISSN 0976 – 6553(Online) Volume 3, Issue 2, July- September (2012), © IAEME

line. The line resistance is taken 0.031ohm per km. The faults have occurred at 15 different
locations.

Fig 1: Simulated radial power system network

In the second case 400V, 15Km six bus eight line non radial power system network as shown
in figure 2 is considered. The length of each line is 15km. The transmission lines are
constructed with 15 identical 3 phase LCC blocks of 1km each. The voltage for the two
generators is assumed to be 11KV and the MVA rating is 10MVA. The rating of the power
transformers are 11KV/440V and 10MVA, each having delta-star connection. The three phase
balanced load are connected at the four load buses. All types of faults are simulated at 1km
intervals in all the eight lines. The voltage signals are monitored from bus 1 only. The line
resistance is taken 0.0585ohm per km.
The fault resistance considered for both the cases is 10ohms and the total simulation time in
80 milliseconds. The sampling frequency is 2000 samples per cycle. All the different types of
faults are generated after 40 milliseconds i.e. after two cycles. The technique developed in this
paper mainly detects the type of faults in the transmission line.

Fig 2: Simulated 6bus-8line non radial power system network

3. FEATURE EXTRACTION USING WAVELET TRANSFORM

Feature extraction can be defined as a unique process that transforms the raw signals from its
original form to a new form to extract suitable information from them. The feature extraction
step is crucial in an automatic classification system. This is because a classifier can operate
reliably only if the features of each event are selected properly. During fault, the amplitude
and frequency of the test signal will change significantly as the system change from normal
state to faulty state. The Shannon entropy will change accordingly. Wavelet transformation
has the ability to analyze different transmission line faults simultaneously in both time and
frequency domains. The wavelet transform is useful in detecting and extracting faulty features
of various types transmission line faults both in radial and non radial network because it is
sensitive to signal irregularities but insensitive to the regular-like signal behavior. The wavelet

96
International Journal of Electrical Engineering and Technology (IJEET), ISSN 0976 –
6545(Print), ISSN 0976 – 6553(Online) Volume 3, Issue 2, July- September (2012), © IAEME

transform can be interpreted as the inner product of the complex conjugate of wavelet function
h* a,τ(t) and the input signal s(t)
WT (a,τ) = ∫ h* a,τ(t) . s (t)dt (1)
Where the wavelet function h* a,τ(t) is defined as
h*a,τ(t) = a¯¹/² h(t-τ/a) (2)
The variable a is the scale parameter of the wavelet function and is proportional to the
reciprocal of the frequency, τ is the translation parameter and h is called the mother
wavelet. Wavelet analysis deals with unsteady signal while entropy expresses information of
the signal. Hence Wavelet entropy can analyze fault signals more efficiently. The feature
extraction is very important in signal processing operations because the rough and large data
sets cause difficulties, when a network is trained. In this study half cycle of pre fault and half
cycle of post fault of the faulty voltage signals of all the three phases are considered to reduce
the data set in size.
Wavelet Packet is employed on all the voltage signals which are originated from EMTP
simulation for obtaining high frequency detail component which gives distinctive features
about the signals. A mother wavelet is a function that oscillates, has finite energy and zero
mean value. Normally, the mother wavelet that resembles the analyzed signal is the best
choice, as this would reduce the number of non-zero Wavelet coefficients. For our project
purpose DB4 is selected as the mother wavelet. DB4 owns a good time resolution providing an
accurate detection of the fast transients induced by faults.
In this paper wavelet coefficients of voltage signals are decomposed at level 1 for
classification of fault types. Entropy values are considered as the input features for
classification of different types of faults (SLG, LL, DLG and LLL) at different locations.
Entropy is a common method in many fields, especially in signal processing applications.
Entropy indicates the amount of information which is stored in observed signal. Wavelet
packet is applied to find the entropy values of different voltage signals at different levels.

4. CLASSIFICATION OF FAULTS USING ARTIFICIAL NEURAL NETWORK

An artificial neural network can be defined as a data processing system consisting of a large
number of simple, highly interconnected processing elements (artificial neurons) in an
architecture inspired by the structure of the cerebral cortex of the human brain. ANN is
powerful in pattern recognition, classification and generalization. Neural networks are useful
for power system applications because they can be trained with off line data.
Probabilistic neural networks (PNN) are a kind of radial basis network suitable for
classification problems. The architecture is feed forward in nature which is similar to back
propagation, but differs in the way that learning occurs. PNN is supervised learning algorithm
but includes no weights in its hidden layer. Instead each hidden node represents an example
vector, with the example acting as the weights to that hidden node. These are not adjusted at
all. Figure 3 illustrates a sample PNN structure.

97
International Journal of Electrical Engineering and Technology (IJEET), ISSN 0976 –
6545(Print), ISSN 0976 – 6553(Online) Volume 3, Issue 2, July- September (2012), © IAEME

Figure 3: PNN Architecture

Basically, PNN consists of an input layer, which represents the input pattern or feature vector.
The input layer is fully interconnected with the hidden layer, which consists of the example
vectors (the training set for the PNN). The actual example vector serves as the weights as
applied to the input layer. Finally, an output layer represents each of the possible classes for
which the input data can be classified. However, the hidden layer is not fully interconnected to
the output layer. The example nodes for a given class connect only to that class's output node
and none other. The other important element of the PNN is the output layer and the
determination of the class for which the input layer fits. This is done through a winner-takes-
all approach. The output class node with the largest activation represents the winning class.
While the class nodes are connected only to the example hidden nodes for their class, the input
feature vector connects to all examples, and therefore influences their activations. Therefore
the sum of the example vector activations determines the class of the input feature vector.
In PNN algorithm, calculating the class-node activations is a simple process. For each class
node, the example vector activations are summed, which are the sum of the products of the
example vector and the input vector. The hidden node activation, shown in the following
equation, is simply the product of the two vectors (E is the example vector, and F is the input
feature vector).
(3)
The class output activations are then defined as:
hi −1
N
2

∑e γ
i =1
Cj = (4)
N
Where N is the total number of example vectors for this class hi is the hidden-node activation,
and γ is a smoothing factor. The smoothing factor is chosen through experimentation. If the
smoothing factor is too large, details can be lost; again if the smoothing factor is too small, the
classifier may not generalize well.
Total 15 sets of data, each set containing entropy values of the pure and ten types of faulty
signals are created for the training and testing of the neural network for the radial and non
radial power system network to identify the fault types. Out of these 15 sets, 7 sets are used
for the training purpose and remaining 8 sets of data are used for the testing purpose for radial
network. Similarly 9 sets are used in the training purpose and remaining 6 sets of data are used
for the testing purpose for non radial network.

98
International Journal of Electrical Engineering and Technology (IJEET), ISSN 0976 –
6545(Print), ISSN 0976 – 6553(Online) Volume 3, Issue 2, July- September (2012), © IAEME

4.1 Training of PNN

The designed PNN model is trained by 77 signals. The size of the input vector for type
identification in single transmission line is 6×77, where 6 are approximation and detail values
of level 1 for all the three phases and 77 comes from 11 types of signals multiplied by 7
different locations. The size of the input vector for type identification in non radial power
system network is 6×792, where 792 comes from 11 types of signals multiplied by 9 different
locations in eight different lines.
Target of PNN for fault classification are 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11 for Pure, SLG-R, SLG-
Y, SLG-B, LL-R-Y, LL-Y-B, LL-B-R, LLG-R-Y, LLG-Y-B, LLG-B-R, LLL faults
respectively.
4.2 Testing of PNN
Pure signal along with ten different types of faults are considered in each set. In radial network
all the 11 signals in eight different locations are used for the testing purpose. Hence 88
different signals are created for radial network to classify the types of faults.
In non radial network similar 11 types of waveforms are considered. These signals are taken in
6 different locations in 8 lines. Hence 48 different locations are made. 11 types of signals
multiplied by 48 different locations give 528 different signals for non radial network for the
testing of the PNN model for classification of faults.

5. RESULTS

After training, the neural network based fault detector and classifier is extensively tested using
independent data sets consisting of fault scenarios not used previously for training the
network. Fault type, fault location and fault resistances are changed for different faults in the
validation/test data set to investigate the effects of these factors on the performance of the
proposed method.
In the case of single transmission line ten different types of faults along with the pure signal
are classified very accurately at eight different locations. Table 1 gives the classifier
performance for each type of fault. The classifier gives 100% accurate results for LLG and
LLL type of faults. In the case of SLG and LL there are some misclassifications. Hence, the
overall accuracy of the classifier is 93.18%.
The results of classification for non radial system are shown in table 2. To investigate the
accuracy of the proposed method in these cases, 100% accurate results are found for SLG and
LLG type of faults. The accuracy rate is slightly decreased for some misclassifications in LL
and LLL faults. In some cases three phase symmetrical faults are classified as line to line
faults. Hence PNN model based fault identifier identifies the fault types with an accuracy of
93.94% in a very fast and effective manner. Fault resistances are varied to identify the fault
types at different lines in non radial power system network. Fault resistances are varied from 0
to 30ohms.

99
International Journal of Electrical Engineering and Technology (IJEET), ISSN 0976 –
6545(Print), ISSN 0976 – 6553(Online) Volume 3, Issue 2, July- September (2012), © IAEME

Table 1: ANN result for fault classification in single transmission line

Type Pure LGR LGY LGB LLRY LLYB LLBR LLGRY LLGYB LLGBR LLL
1 2 3 4 5 6 7 8 9 10 11
1 8 0 0 0 0 0 0 0 0 0 0
2 0 8 0 0 0 0 0 0 0 0 0
3 0 0 8 0 0 0 0 0 0 0 0
4 0 0 1 5 0 0 1 0 1 0 0
5 0 0 0 0 8 0 0 0 0 0 0
6 0 0 0 0 0 7 0 0 1 0 0
7 0 2 0 0 0 0 6 0 0 0 0
8 0 0 0 0 0 0 0 8 0 0 0
9 0 0 0 0 0 0 0 0 8 0 0
10 0 0 0 0 0 0 0 0 0 8 0
11 0 0 0 0 0 0 0 0 0 0 8
Overall accuracy 93.18%

Table2: ANN result for fault classification in non radial system

Type Pure LGR LGY LGB LLRY LLYB LLBR LLGRY LLG YB LLGB LLL
1 2 3 4 5 6 7 8 9 R 11
1 48 0 0 0 0 0 0 0 0 10
0 0
2 0 48 0 0 0 0 0 0 0 0 0
3 0 0 48 0 0 0 0 0 0 0 0
4 0 0 0 48 0 0 0 0 0 0 0
5 0 0 0 0 35 5 4 0 0 0 4
6 0 0 0 0 0 48 0 0 0 0 0
7 1 0 0 0 0 3 44 0 0 0 0
8 0 0 0 0 0 0 0 48 0 0 0
9 0 0 0 0 0 0 0 0 48 0 0
10 0 0 0 0 0 0 0 0 0 48 0
11 1 0 0 0 1 4 8 0 0 0 34
Overall accuracy 93.94%

6. CONCLUSIONS

The purpose of this paper is to find an effective technique suitable for the fault classification
of power system transmission lines. The proposed method uses wavelet packet decomposition
which provides more features about the signal than classical wavelet. After wavelet packet
decomposition of voltage signals up to three decomposition levels, the entropy values are
calculated only for detail level 1 and 3 for each faulty voltage waveforms. Therefore the size
of the feature vectors is reduced considerably. The features obtained by this way are used as
the inputs of ANN for determination of all types of faults along with affected phases within
half cycle after the occurrence of the faults. The results presented in this paper confirm the
possibility of developing an accurate fault classification scheme that may aid the development
of reliable transient-based protection schemes. Further investigations are being carried out to
confirm the robustness of the performance under changes in the network configuration. The
proposed scheme is easily comprehensible, deterministic and is feasible for practical
implementation.

100
International Journal of Electrical Engineering and Technology (IJEET), ISSN 0976 –
6545(Print), ISSN 0976 – 6553(Online) Volume 3, Issue 2, July- September (2012), © IAEME

ACKNOWLEDGEMENT

The authors acknowledge the financial support given by Department of Science and
Technology (DST), Govt. of India for sponsoring this research to Dr. Sudipta Nath.

REFERENCES

[1] S. El. Safty & A. El-Zonkoly (2009), “Applying wavelet entropy principle in fault
classification” Electrical Power and Energy Systems 31 pp.604-607
[2] Pradhan A. K. et Al. (Oct. 2004), “Wavelet-fuzzy Combined Approach for Fault
Classification of a Series-Compensated Transmission Line” IEEE Trans. on Power Delivery,
Vol. 19, No. 4.
[3] Jung. H et. al. (2007), “Novel Technique for Fault location estimation on parallel
transmission lines using wavelet” Electrical Power and energy Systems 29 pp.76-82
[4] Kim. K.H. et al. (2007), “Wavelet and Neuro-Fuzzy Fault Location for Combined
Transmission Systems” Electrical Power and Energy Systems 29 pp. 445-454
[5] Chunju. F et Al. (2007), “Application of Wavelet Fuzzy Neural Network in Location
Single Line to Ground Fault (SLG) in Distribution Lines” Electric Power and Energy Systems
29 pp. 497-503
[6] Eisa. A. Amir. A. & Ramar. K, (2010) “Accurate one-end Fault location for Overhead
Transmission Lines in Interconnected power system” Electrical Power and Energy Systems 32
pp. 383-389
[7] Jain. A., Thoke. A. S., Patel. R. N. (2008) “Fault Classification of Double Circuit
Transmission Line using Artificial Neural Network” International Journal of Electrical and
Electronics Engineering 1:4
[8] Salim. R.H et al. (April 2009), “Extended fault-Location Formulation for Power
distribution Systems” IEEE Trans. on Power Delivery, Vol.24, No.2
[9] Bingyong Zan, (2006) “A Novel distributed approach to Robust Fault Detection and
Identification” IEEE Trans. on Power Delivery,
[10] Borghetti. A et al. (2006), “On the Use of Continuous-Wavelet Transform for Fault
Location in Distribution Power Systems” Electrical Power and Energy Systems 28 pp. 608-
617
[11] Fernando H. M. & Abur. A. (October 1998), “Fault Location Estimation on Parallel
Transmission Lines using Wavelet” IEEE Trans. on Power Delivery, Vol. 13, No 4.
[12] Saleh A. & Al-jufout (2008), “Improved algorithm for Fault Simulation by hypothetical
Stub moving along a medium-length transmission Line” ISSN 1450-216X Vol. 23, No 3 pp.
400-405
[13] Ekici S., Yildirim S., Poyraz M., (2008) “Entropy and energy based feature extraction for
locating fault on transmission lines by using neural network and wavelet packet
decomposition” Expert Systems with Applications 34 2937-2944
[14] Chanda D., Kishore N. K., Sinha. A, (October 2004), “A wavelet multi-resolution based
Analysis for Location of the Point of Strike of a Lightning over voltage on a Transmission
Line” IEEE Trans. on Power Delivery.

101
International Journal of Electrical Engineering and Technology (IJEET), ISSN 0976 –
6545(Print), ISSN 0976 – 6553(Online) Volume 3, Issue 2, July- September (2012), © IAEME

AUTHORS PROFILE

Soumyadip Jana obtained his B.Tech. & M. Tech. from Bengal College of Engineering &
Technology,Durgapur and Netaji Subhash Engineering College, Garia, Kolkata respectively,
in India . He is currently Assistant Professor in the Department of Electrical Engineering,
Techno India, Salt Lake, West Bengal, India. His research interest include, power systems and
application of artificial intelligence in power quality

Sudipta Nath obtained her B.E., M.Tech. and PhD in Electrical Engineering from Regional
Engineering College, Durgapur, University of Calcutta and Bengal Engineering and Science
University, Shibpur, respectively, in India. She is currently holding the position of Professor,
Department of Electrical Engineering, Netaji Subhash Engineering College, West Bengal,
India. Her research interests include, power systems and digital signal processing. She has
number of research papers published in national and international journals.

Aritra Dasgupta obtained his B.Tech. and M. Tech. from Netaji Subhash Engineering
College. He is currently Assistant Professor in the Department of Applied Electronics and
Instrumentation Engineering, Netaji Subhash Engineering College, West Bengal, India. His
research interest includes application of artificial intelligence in power quality.

102