International Journal of Scientific & Engineering Research Volume 9, Issue 2, February-2018 678

ISSN 2229-5518

FAULT IDENTIFICATION SYSTEM FOR

ELECTRIC POWER TRANSMISSION LINES
USING ARTIFICIAL NEURAL NETWORKS
Mbamaluikem, Peter O., Aderemi, Oluwaseun S., Awelewa, Ayokunle A.

Abstract— Electric power transmission line faults hinder the continuity of electric power supplied and increase the system downtime
thereby increasing the loss of electric power transmitted. Early fault detection and classification leads to prompt clearance of faults
with an attendant effect of improved reliability and efficiency of the power system network. In view of this, this paper develops an arti-
ficial neural network (ANN)-based detector and classifier to indicate and classify respectively a fault on Nigeria 33-kV electric power
transmission lines. The transmission lines are modeled in Simulink using SimPowerSystems toolbox in MATLAB. Fault simulations are
carried out, and the resulting instantaneous values of voltages and currents are used to develop the proposed fault identification sys-
tem using multilayer perceptron feedforward artificial neural networks with backpropagation algorithm. Results are presented to vali-
date the effectiveness and efficiency of the developed identification system for detecting and classifying faults. The Mean Square Error

IJSER
(MSE), linear regression and the confusion matrix are used as performance evaluators for the system. The ANN-based identification
system achieved MSE of 4.77399e-10 and an accuracy of 100% for fault detection. This indicates that the performance of the developed
ANN-based identification system is highly satisfactory and may be practically implemented on the Nigeria transmission lines.

Index Terms— Artificial neural networks, SimPowerSystems, Backpropagation algorithm, Transmission lines, Fault detector, Fault classifier

——————————  ——————————

1 INTRODUCTION
Overhead transmission lines are one of the primary elements in sudden disturbances that can cause an abnormal flow of current
an electric power system transmission network. The electric pow- in the line conductors. Faults are widely classified as series faults
er transmission line networks consist of interconnected electrical and shunt faults. The most occurred and severe type of fault is the
conductors carried on a support which convey electrical power shunt faults and its analysis is used in the design of protective and
from different generating power station to the grid and from the relaying equipment [3]. These shunt faults may be caused by
grid to different substations at varying degree of voltages. In Ni- lightning strokes, trees falling across lines, vehicle colliding with
geria, there are four levels of transmission voltages, i.e., 330-kV, the poles, trees growing up to the transmission lines, birds short
132-kV, 33-kV and 11-kV and these voltage levels use different circuiting the lines, etc. When a fault occurs on an electric trans-
sizes and/or types of conductors in conveying electrical power mission line, it is imperative to detect and classify it to quicken
[1]. Of these levels of voltages, this paper considers the 33-kV the necessary arrangement towards restoring the power hence
transmission network, which is characterized by very lengthy reducing the down time of lack of power to consumers which is
transmission lines; often it passes through bushes and is more part of the main goal of transmission line protection [4].
close to the environment compared to 330-kV and 132-kV trans- Therefore, to reduce the downtime of lack of electric power to
mission lines. This makes it more prone to faults hence the possi- consumers, a reliable intelligent fault detection and classification
bility of experiencing faults on 33-kV transmission lines is gener- system is required to improve identification of the faults on
ally higher than in others [2]. Faults on transmission lines are transmission lines. This effort is worthwhile considering the eco-
IJSER © 2018
http://www.ijser.org
International Journal of Scientific & Engineering Research Volume 9, Issue 2, February-2018 679
ISSN 2229-5518
nomic benefits of quick detection and classification of faults [5]. backpropagation algorithm is employed in the training of two
The pattern recognition technique is useful in selecting the faulty different ANN configurations which are used for fault detection
and healthy transmission lines in electrical power system trans- and classification, respectively.
mission network. Its ability is used to differentiate among the
three lines which one is experiencing a fault. The ANNs are very 2. Artificial Neural Network
powerful in fault identification by pattern recognition. To this Artificial Neural Network (ANN) is a massively interconnected
end, this paper uses artificial neural networks to develop fault parallel computational system modeled in the way biological
identification system to realize a complete fault detection and neurons work. They are made of several processing elements
classification for all types of shunt faults on the 33-kV Nigeria joined together in a particular way in order to achieve a particular
transmission line. outcome [16],[17]. This makes them more efficient for real-world
Gowrishanka, et al [6] combined the capability of discrete wave- applications. In addition, its capability to process information in a
let Transform and artificial neural network to develop a fault de- parallel distributed manner, solve problems that are naturally
tector and classifier in transmission lines. Leite, et al [7] devel- nonlinear without having prior knowledge functions relating the
oped a new technique for the detection and location of high-speed problem variables and its capability in handling incomplete in-
faults using neural networks. The review of different approach formation and corrupt data, gave it an edge over the other artifi-
was previewed in [8]. Fault detection and classification for

IJSER
cial intelligence in fault identification in transmission lines [18].
transmission line protection system using artificial neural net- During training, an artificial neural network learns to produce a
work was proposed in [9]. Lv and Zhang, [10] employed desired output based on a given input data. Hence, a training
feedforward artificial neural network with backpropagation algo- dataset is needed to train any artificial neural network to perform
rithm for fault classification on transmission line of 10kV rural a particular operation [19]. The training of the network is accom-
power grid. A powerful fault classifier and locator using artificial plished by sequentially applying input vectors while adjusting
neural network with supervised learning was proposed in [11]. network weights accordingly and passing the same through an
Silva, et al. [12] employed Wavelet Transform and ANN for de- activation function. The network weight converges gradually as
tection and classification of faults in power transmission lines. A the adjustment of different weights progress to values that will
fault detection and location system for high-speed protection in enable each input vector to produce the target [20]. Figure 1 is a
extra high voltage transmission lines using feedforward neural simple structure of a feedforward multilayer perceptron artificial
network with the backpropagation algorithm based on supervised
learning was applied by [13]. Seema, et al. [14] used ANN with
gradient descent backpropagation algorithm to implement an in-
telligent fault identification system. The system realized is capa-
ble of detecting only single line-to-ground and double line-to-
ground faults. A multilayer perceptron backpropagation neural
network combined with Neuro shell-2 software is proposed in
[15] to develop a fault detector, classifier, and locator for a neural network.
transmission line. In this paper, the modeling and simulation of a
33-kV practical electric power transmission system is performed
Figure 1: Simple Structure of a feedforward Multilayer
in MATLAB 2015a environment. The results of the simulations
Perceptron ANN
formed the input database for the proposed ANN. The multilayer
perceptron feedforward artificial neural network with

IJSER © 2018
http://www.ijser.org
International Journal of Scientific & Engineering Research Volume 9, Issue 2, February-2018 680
ISSN 2229-5518
2.1 ANN Implementation Steps [0.18446 0.39072]
The development of the fault identification system using ANN Positive- and zero-sequence inductances (H/km):
follows these procedures: [0.0010981 0.0024668]
1 Transmission line simulation, sampling and recording of Positive- and zero-sequence capacitances (F/km):
fault currents and voltages. [1.0865e-08 6.6177e-09]
2 Preprocessing of recorded fault currents and voltages Fault Starting = 0.020 seconds
3 Selection of the ANN configuration Duration in fault = 0.03 Seconds
4 Training of the ANN Type Conductor = ACSR
5 Performance evaluation and validation of the trained
A. Simulation and Preprocessing of data
ANN to ascertain its capability to generalize.
The system model is first set up in Simulink, and then the three-

3. Power System Transmission Line Modeling phase V-I measurement block is used to measure the instantane-
ous values of voltages and currents signals. The three section pi-
The snapshot of the Simulink model of 33-kV transmission line
section line 1 and 2 together is 140 km long. Ten different shunt
of Isolo transmitting station is shown in Figure 2. It consists of
faults plus no fault condition are simulated between 1 km and
140 km length of 33-kV transmission line. A source is located at
140 km at a step of 2 km. The values of resistances and fault in-

IJSER
one end and a load attached at the other end. The transmission
ception angle used are shown in Table 1. These faults are simu-
lines are modeled in Simulink/MATLAB 2015a environment
lated one after another and the three phase voltage and current
waveforms generated are sampled at a frequency of 1.5 kHz.
Hence, there are thirty (30) samples per cycle. These samples are
preprocessed to obtain a suitable input data set for the ANN [21].
In all, a total of 6 x 6,160 sample data set is obtained for this
study.

Table 1: Parameter Values used in Generating the Training

and Test Dataset
using Pi model. Training Dataset
Fault inception angle (0) 30, 60
Figure 2: Snapshot of the modeled 33-kV Isolo transmission line Fault Resistance (Ohms) 0.25, 0.5, 0.75, 5, 10, 20, 30
in Simulink/Matlab and 50
The sampling frequency of the model is 1.5 kHz. The simulation Test Dataset
time is set to 0.02 second. In the model, the current and voltage Fault Location (km) 8, 16, 24, …, 138
measurements are carried out at bus B1. The instantaneous values Fault Resistance (Ohms) 15, 25
0
of voltages and currents obtained are used for fault detection and Fault inception angle ( ) 20, 90
classification. Ten different faults scenario were simulated at var-
ying distance and resistance values. The transmission line charac- 4. The Proposed Fault Identification System
teristics and parameters used for the model are: The fault identification system being proposed is designed to
Line Length = 140 km detect the presence of a fault and afterward classify the fault. The
Positive- and zero-sequence resistances (Ohms/km): developed system has two stages – the detection stage and the
IJSER © 2018
http://www.ijser.org
International Journal of Scientific & Engineering Research Volume 9, Issue 2, February-2018 681
ISSN 2229-5518
classification stage. The Figure 3 shows the modular block repre- data which means 560 samples for each of the faults considered
senting the proposed system while the stages involved in the de- and No-fault case. This invariably forms an input-output pattern
velopmental process are as follows: of set of six (6) inputs and an output pair. Several configurations

1. Acquisition and Pre-Processing of input Data set of the ANN with varying number of neurons are extensively

2. Preparation of appropriate target data set suitable for the trained and the results of the structure that gives the best satisfac-

ANN to learn tory performance are presented. The performance MSE, confu-

3. Choose ANN configuration sion matrix, regression plot and the generalization capability are

4. Train the ANN used as performance indicators. Table 2 shows the truth table of

5. Calculate the performance MSE the fault state of the lines.

6. Evaluate the trained ANN using the Performance MSE,

confusion matrix, Regression Plot and a new set of data
outside the one used for training.
7. Generate and save a Simulink Model for the best per-
formed ANN based on the test matrix.

IJSER
Fault Classifier
Sampling / Fault De-
I, V
Normalization tector
0: No-Fault
1: Fault

Display

Figure 3: The modular block diagram of ANNFD

Table 2: The Target Truth Table for the ANNFD
Fault Type
4.1 Fault Detector A B C A- A- B- A A B A-
No-
The proposed ANN-Based fault detector (ANNFD) is designed - - - B- C- C- - - - B-
Fault
using the preprocessed instantaneous current and voltage data set G G G G G G B C C C
generated to indicate the absence or presence of a fault on the 33-
Line
kV Nigeria electric transmission lines [22]. In consequence, if the 1 1 1 1 1 1 1 1 1 1 0
Status
ANNFD detects a fault, it activates the corresponding ANN-
Based fault classifier (ANNFC) for classification of the detected
fault. The ANNFD uses one-terminal line datum extracted at bus Results and Discussion
B1and takes in six (6) inputs at a time which are the preprocessed The ANN structure with 6-5-1 configuration having activation
instantaneous voltages and currents for the ten (10) different fault function combination of purelin/logsig/tansig for the input layer,
cases and no-fault case simulated. A total of six thousand, one one hidden layers and output layer respectively gave the best
hundred and sixty (6,160) input data set is extracted, scaled and satisfactory performance among several structures considered.
used as input data for training the ANNFD. Hence, the training Furthermore, the developed ANNFD confusion matrix, validation
set consists of 6 x 6,160 input data set and 1 x 6,160 target output performance plot, and linear regression plot are presented in Fig-
IJSER © 2018
http://www.ijser.org
International Journal of Scientific & Engineering Research Volume 9, Issue 2, February-2018 682
ISSN 2229-5518
ure 7, Figure 8, and Figure 9, respectively and are used to ascer-
tain the capability of the ANNFD to generalize [6].
It can be seen from the blue square box of Figure 7 that the per-
formance of the developed fault identification system is 100%,
showing that there is no confusion in detecting all the faults test-
ed. In addition, the performances plot (Figure 8), shows that the
best validation performance MSE is 4.77389e-10 which is below
the preset MSE goal of 0.0001 and finally, the linear regression
fit (Figure 9) shows the relationship between the actual neural
network’s outputs and the desired targets. However, from Figure
9, the correlation coefficient, R is 0.99789 which is very close to
the ideal value (1), thus indicating an excellent correlation.
Therefore, the performance of the developed system is said to be
satisfactory and the system can be said to have very good train-
Figure 7: Confusion Matrix for ANN-Based FD with 6-14-4 configuration ing, validation, and testing. Hence, the developed ANNFD can be
deployed to the 33-kV Nigeria transmission line.

IJSER
4.2. Fault Classification

The process of developing artificial neural network based fault

classifier (ANNFC) employed for ANNFD. The ANN takes in a
set of 6 x 6,160 as input data set as earlier emphasized and 4 x
6,160 target output which in turn forms an input-output pattern of
six (6) inputs and four (4) output pair. The four output target of
the neural network corresponds to the fault state of each of the
three lines and the ground line. To this end, the various permuta-
tion of the fault state of the lines with the ground is shown in Ta-
Figure 8: Performance Plot for one hidden layers with 6-14-4 configuration
ble 3. Moreover, it is expected that the developed system should
be able to accurately classify the ten possible categories of faults
and the No-fault case. The training data set holds total of 6,160
inputs data set and a corresponding output target i.e. 560 for each
of the fault type and No-fault condition. The MLP feedforward
Back-propagation neural networks with combinations of different
hidden layers, number of neurons in each hidden layer and trans-
fer function are thoroughly and exhaustively investigated. As
before, the mean square error of the trained network, confusion
matrix, linear regression and receivers operating characteristics
are used as performance matrix to test the proposed system.

Figure 9: The ROC for ANNFD with 6-14-4 Configuration.

IJSER © 2018
http://www.ijser.org
 International Journal of Scientific & Engineering Research Volume 9, Issue 2, February-2018 683
ISSN 2229-5518

Table 3: The Target Truth Table for the ANNFC

Fault Type
A B C A- A- B- A A B A-
Line No-
- - - B- C- C- - - - B-
Fault
G G G G G G B C C C
A 1 0 0 1 1 0 1 1 0 1 0
B 0 1 0 1 0 1 1 0 1 1 0
C 0 0 1 0 1 1 0 1 1 1 0
G 1 1 1 1 1 1 0 0 0 0 0

Figure 10: Performance Plot for one hidden layer with 6-16-4 configuration
A. Results and Discussion

After training several ANN structures with different configura-

tions, the ANN structure with 6-16-4 configuration having activa-

IJSER
tion function combination of purelin/logsig/tansig for the input
layer, one hidden layer and output layer respectively gave the
best satisfactory performance. Figure 10 – Figure 13 show the
performance matrix used in determining the performance of the
developed ANNFC. From Figure 10, it can be said that the
ANNFC is efficient since the testing and the validation curves
have similar characteristics and the validation performance MSE
is 0.000301203 which is below the preset MSE of 0.001. Also, an
accuracy of 94.5% as seen in Figure 11 and the correlation coef- Figure 11: Confusion Matrix for ANNFC with 6-16-4 configuration
ficient of 0.99911(Figure 12) which shows how much the net-
work output was able to track the target further proves the satis-
factory performance of the developed ANNFC. From the ROC
plot (Figure 13) which expresses the percentage of true positivity
and false positivity, it can be seen that all the line are well fitted
in the upper-left corner which expresses a 100% true positivity
and 0 percent false positivity of classification. Based on the re-
sults, this study concluded that the ANNFC developed can effec-
tively and efficiently differentiate among all the ten possible
types of faults considered and No-fault condition on the 33-kV
Nigeria transmission line.

Figure 12: The Regression Plot for the ANNFC with 6-16-4 Configuration

IJSER © 2018
http://www.ijser.org
International Journal of Scientific & Engineering Research Volume 9, Issue 2, February-2018 684
ISSN 2229-5518
Dr. Awelewa, Ayokunle A., is a Senior Lecturer in the Depart-
ment of Electrical and Information Engineering, Covenant Uni-
versity, Canaanland, KM 10 Idiroko Road, PMB 1023, Ota, Ogun
State, Nigeria. Email id: ayokunle.awelewa@covenant-
university.edu.ng

References
[1] W. P. Davis, Analysis Of Faults In Overhead
Transmission Lines, (California State University,
Sacramento, 2012).
[2] F. O. Akpojedje, M. E. Onogbotsere, E. C. Mormah and
P. E. Onogbotsere, "A Comprehensive Review of Nigeria
Figure 13: The ROC for ANNFC with 6-16-4 Configuration
Electric Power Transmission Issues and Rural
Electrification Challenges," International Journal of
Conclusions
Engineering Trends and Technology (IJETT), vol. 31, no.
The longer faults stay on a transmission line, the more damage it 1, 2016.

IJSER
is likely to cause to the transmission network which will eventu- [3] Gupta, J. B. (2008). A Course in Power Systems. New
ally lead to loss of economy. To this end, artificial neural net- Delhi: S. K. Kataria & Sons.
works have been employed in this paper in designing a fault iden-
tification system for fault identification and classification on the [4] S. B. Ayyagari, Artificial Neural Network Based Fault

33-kV Nigeria electric transmission line. The developed system Location for Transmission Lines, (University of

uses as inputs data to the ANN the preprocessed instantaneous Kentucky Master's Theses, 2011). Paper 657.

values of voltages and currents sampled at a frequency of 1.5 http://uknowledge.uky.edu/gradschool_theses/65.

kHz. Moreover, all the single line-to-ground faults, double line- [5] P. O. Mbamaluikem, A. A. Awelewa, I. S. Samuel, "An

to-ground faults, and line-to-line faults tested have been accurate- Artificial Neural Network-Based Intelligent Fault

ly detected and classified. More so, results presented for the ten Classification System for the 33-kV Nigeria

(10) different fault cases analyzed shows that the developed iden- Transmission Line," International Journal of Applied

tification system is efficient and reliable enough to be imple- Engineering Research vol. 13, no. 2, pp. 1274-1285,

mented on the 33-kV Nigeria electric power transmission line. 2018.

[6] M. Gowrishankar, P. Nagaveni, and P. Balakrishnan,

“Transmission Line Fault Detection and Classification
Mbamaluikem, Peter O., is a Postgraduate Student in the De- Using Discrete Wavelet Transform and Artificial Neural
partment of Electrical and Information Engineering, Covenant Network,” Middle-East J. Sci. Res., vol. 24, no. 4, pp.
University, Canaanland, KM 10 Idiroko Road, PMB 1023, Ota, 1112–1121, 2016.
Ogun State, Nigeria. Email id: [email protected]
Aderemi, Oluwaseun S., is a Postgraduate Student in the De- [7] D. F. Leite, M. B. Hell, P. Costa, and F. Gomide, “Real-

partment of Electrical and Information Engineering, Covenant time fault diagnosis of nonlinear systems,” Nonlinear

University, Canaanland, KM 10 Idiroko Road, PMB 1023, Ota, Analysis, Theory, Methods and Applications, vol. 71, no.

Ogun State, Nigeria. Email id: [email protected] 12, 2009.

IJSER © 2018
http://www.ijser.org
International Journal of Scientific & Engineering Research Volume 9, Issue 2, February-2018 685
ISSN 2229-5518
[8] M. Tarafdar Haque and A. Kashtiban, “Application of Artificial Neural Network, International Journal of
Neural Networks in Power Systems; A Review,” Research in Management, Science & Technology 3(4),
International Journal of Electrical, Computer, Energetic, pp. 87–95, 2015
Electronic and Communication Engineering Vol:1, no. 6,
[17] R. C. Chakraborty, Fundamentals of Neural Network:
2007.
Soft Computing, 2010.
[9] E. P. Thwe, “Fault Detection and Classification for
[18] J. C. S. Souza, M. a. P. Rodrigues, M. T. Schilling, and
Transmission Line Protection System Using Artificial
M. B. Do Coutto Filho, “Fault location in electrical
Neural Network,” Journal of Electricaland Electronics
power systems using intelligent systems techniques,”
Engineering, vol. 4, no. 5, p. 89, 2016.
IEEE Trans. Power Deliv., vol. 16, no. 1, pp. 59–67,
[10] C. Lv and S. Zhang, “Fault classification on transmission 2001.
line of 10kV rural power grid.” 4th International
[19] K. S. Swarup and H. S. Chandrasekharaiah, “Fault
Conference on Sensors, Measurement and Intelligent
detection and diagnosis of power systems using artificial
Materials (ICSMIM 2015). Published by Atlantis Press,
neural networks,” Proc. First Int. Forum Appl. Neural
2016.
Networks to Power System, pp. 3442–3446, 1991.

IJSER
[11] S. K. Kumar, M. Swamy, and V. Venkatesh, "Artificial
[20] A. Yadav and Y. Dash, “An Overview of Transmission
Neural Network Based Method for Location and
Line Protection by Artificial Neural Network: Fault
Classification of Faults on a Transmission Lines,"
Detection, Fault Classification, Fault Location, and Fault
International Journal of Scientific and Research
Direction Discrimination.” Advances in Artificial Neural
Publications, Vol. 4, no. 6, 2014.
Systems. Hindawi publishing corporation, 2014,
[12] K. M. Silva, B. A. Souza, and N. S. D. Brito, “Fault http://dx.doi.org/10.1155/2014/230382
Detection and Classification in Transmission Lines Based
[21] M. Jamil, S. K. Sharma, and R. Singh, “Fault detection
on Wavelet Transform and ANN,” IEEE Transactions on
and classification in electrical power transmission system
Power Delivery, vol. 21, no. 4, pp. 2058–2063, 2006..
using artificial neural network,” SpringerOpen Journal,
[13] T. Bouthiba, “Fault Location in EHV Transmission Lines 2015.
Using Artificial Neural Networks,” International
Journalof Applied Math. Comput. Sci., vol. 14, no. 1, pp. [22] P. O. Mbamaluikem, A. A. Awelewa, I. S. Samuel,

69–78, 2004. “Artificial Neural Networks for Intelligent Fault Location

[14] S. Seema, K. R. Mamatha and S. Thejaswini, "Intelligent on the 33-Kv Nigeria Transmission Line,” International

Fault Identification System for Transmission Lines Using Journal of Engineering Trends and Technology (IJETT),

Artificial Neural Network," IOSR Journal of Computer vol. 54, no. 3, pp. 147-155, 2017.

Engineering (IOSR-JCE), vol. 16, no. 1, pp. 23-31, 2014.

[15] E. Bashier and M. Tayeb, “Faults Detection in Power

Systems Using Artificial Neural Network,” American
Journal of Engineering Research, vol. 2, pp. 2320–847.
[16] V. C. Ogboh and T. C. Madueme, “Investigation of Faults
on the Nigerian Power System Transmission Line Using
IJSER © 2018
http://www.ijser.org