ANN Based Fault Detection & Classification in

Power System Transmission line

Md. Rafayel Bishal Sabbir Ahmed Nur Mohammad Molla
Dept. of Electrical and Electronic Dept. of Electrical and Electronic Dept. of Electrical and Electronic
Engineering Engineering Engineering
Rajshahi Universiy of Engineering & Khulna Universiy of Engineering & Khulna Universiy of Engineering &
Technology (RUET) Technology (KUET) Technology (KUET)
Rajshahi-6204, Bangladesh Khulna-9203, Bangladesh Khulna-9203, Bangladesh
rafayelbishal141080@gmail.com ahmed1403017@stud.kuet.ac.bd nurmohammad.kuet@gmail.com

Kazi Mohammad Mamun Asfaqur Rahman Md. Abdullah Al Hysam

Dept. of Electrical and Electronic Dept. of Electrical and Electronic Dept. of Electrical and Electronic
Engineering Engineering Engineering
Bangladesh Universiy of Engineering & Islamic University of Technology (IUT) Islamic University of Technology (IUT)
Technology (BUET) Dhaka-1704, Bangladesh Dhaka-1704, Bangladesh
Dhaka-1205, Bangladesh asfaqurrahman@iut-dhaka.edu aahysam@iut-dhaka.edu
kazi.mamun06115@gmail.com

Abstract—The development of an artificial neural network high fault impedance, protective relays are not able to
to analyze power system faults is currently under way. classify the fault. Moreover, their fault clearing time is also
However, the produced system is unsatisfactory because it is higher, making it approximately (20ms-60ms) [2]. For fault
prone to failure and takes longer to detect faults. The ultimate
goal of this project is to create an artificial neural network
detection and classification, various types of algorithms have
system that can identify and classify all forms of faults in power been developed over the years. Expert system, Fuzzy logic
grid lines soon after they occur. In principle, three-phase systems and the artificial neural network are the updated
currents are used as inputs, and the suggested approach takes algorithm for the power system analysis [3]. Among these
into account all sorts of defects. Following that, an ANN available techniques, Artificial Neural Networks (ANN) has
training network for a power system was constructed in been considered the most promising one because of the
MATLAB/Simulink software, and the ANN network was
trained using a feed-forward approach. Finally, the simulation
higher efficiency compared than others [3]. The artificial
shows that when a faulat occurs in a transmission line, the ANN neural network has also the features of powerful pattern
accurately diagnoses the fault and minimizes any potential recognition, classification, generalization and largely power
damage by efficiently identifying faults in overhead system application [4].
transmission systems. The invention of the artificial neural network is based on
Keywords—Transmission line, Fault detection, Fault the biological ideas along with control applications and this
classification, ANN.
network is mainly an interconnected group of neurons [5]. A
neuron is mainly a data-processing unit which is the
I. INTRODUCTION
fundamental element to operate a neural network. The model
In the power system there are numerous components are of a neuron is shown in the block diagram of fig. 1.
connected from production to distribution of electric power.
In other words, the electrical power system can be divided
into mainly three parts which are generation side,
transmission side and distribution side. The highest number
of faults occur in the transmission sides of the power system
as the overhead lines are expanded over long distances. There
are many reasons for which the transmission lines face the
faults such as lighting, vehicles, storm, transmission tower
breaking. Moreover, animals or birds short-circuiting and
tree breaking are the main reasons for a fault occurring in
overhead lines. A fault is simply known as when the
insulation fails of the line and then the stable power system
becomes unstable. Faults can be classified into two types on Fig. 1. Model of a neuron.
overhead transmission lines which have been designated as From the above fig. 1. The hidden layers neuron number
the symmetrical fault and unsymmetrical fault. Symmetrical mainly determines the problem-solving quality and the fault
and unsymmetrical faults are also called shunt faults (A-G, clearing time directly depends on the hidden layer number
C-G, B-G, A-B-G, B-C-G, A-C-G, A-C, A-B, B-C, A-B-C, [6]. The number of the hidden layer is required to be minimal
A-B-C-G) [1]. If a fault occurs on an overhead transmission to decrease the fault clearing time. In ANN there are three
line, it should be cleared as quickly as possible and then main uses which are classification, noise reduction and the
classified which types of faults have occurred. Protective Extrapolation [7]. Feed-forward neural network is used in
relays are used to sense the fault to trip the circuit breaker. this research and in feed-forward network, the information
However, due to offsetting in the short circuit current and travels in only one direction from input to output. The

978-1-6654-2132-4/21/$31.00 ©2021 IEEE

meanings of the training of artificial neural network comes and it is to be mentioned that the number of inputs need not
down the term weights which are the main element of the to be equal of the outputs. The ANN model will just perform
artificial neural network. The main focus of feed forward according to the trained data set. A block diagram of the basic
system is to minimize the error by adjusting the weights. The principle to classify the fault through the ANN network is
weights adjustment occurs at the time when the network is shown in the following fig. 3.
presented with the input data and their corresponding target Input Output
values. The input values are the three-phase currents. Neural
network trained by off line data and it is useful for power
Ia A-G
system applications. The neural network for this research is
mainly concerned with optimization of fault clearing time B-G
and accurate fault classification[8]. This ANN model can be Ib C-G
developed by MATLAB and Simulink as a neural network
A-B
toolbox is available in MATLAB software [9].

ANN Classifier
RMS Value
Continuous research is going on this field to combine the Ic A-C

relay and the ANN network for fault detection [10]. The B-C
ANN system to find the fault location precisely is also a Va A-B-G
developing research area [11]. Eisa Bashier M Tayeb
A-C-G
proposed a neural network that detects fault on power
transmission line it only discussed only fault detection part, Vb B-C-G
not the classification part [12]. Ahmed Sabrie Altaie has also
A-B-C
shown his research about the fault detection and Vc
classification using the ANN network but his fault clearing A-B-C-G
time was 22ms which should be lower for the safety of the
components [13]. The above-mentioned problems are Fig. 3. Fault classification block diagram.
minimized in the proposed work and the fault clearing time Fig. 3 shows the 11 possible shunt faults and the 3-φ
is also less than 10ms for some of the cases. RMS voltages & currents are the input of the fault classifier
This paper is structured as follows: Section II illustrates neural network. Then the fault classifier neural network
the fault detection & classification; Section III shows the delivers the types of shunt faults. The most desired outcome
ANN data training; Section IV describes the Simulation and of the ANN system is that it not only classifies the fault but
result and finally, the conclusion is well organized in Section also identifies the faulted phase. Suppose for single line to
V. ground fault it specifies the faulted phases (A/B/C) and the
double line to ground fault it also detects the faulted phases
II. FAULT DETECTION AND CLASSIFICATION (A-B/ B-C/ C-A).

To detect a fault two steps need to be followed, the first The ANN model is mandatory to classify and detect
one is to identify the fault in times of high fault current flows faulted phase and its working principle is based on a truth
through the line, and the second one is to the operation of table shown in the following table 1.
circuit breaker to clear the fault. Moreover, after fault TABLE I. TRUTH TABLE OF FAULT CLASSIFICATION
detection on the transmission line, it should immediately Fault type Phase A PhaseB Phase C Ground
identify the type of fault. The ANN model will be used to
Normal 0 0 0 0
classify the transmission line fault as shown in fig. 2.
A-G 1 0 0 1
B-G 0 1 0 1
C-G 0 0 1 1
A-B 1 1 0 0
A-C 1 0 1 0
B-C 0 1 1 0
A-B-G 1 1 0 1
A-C-G 1 0 1 1
B-C-G 0 1 1 1
A-B-C 1 1 1 0
A-B-C-G 1 1 1 1
TABLE I. shows that there are 4 outputs in the ANN model
Fig. 2. Neural network representation to classify the fault
and three of them is denoted as phase corresponding to the
In the above Fig. 2. It is seen that there are three layers in the fault condition and the remaining one is known as
neural network to classify the fault which are input, hidden ground.Hence the outputs are either a 0 or 1 for the absenceor
and output layers. In the output section for the fault classifier existence of a fault on a related line (A, B, C or G where A
ANN phase A, phase C, phase B and ground are specified B and C means the three phases of the transmission line and
G denotes the ground). The ANN model can successfully by 1050 samples of data set and the fault classifier neural
classify the fault also it can detect the faulted phase through network trained by 21000 samples of data set.
this truth table.
IV. SIMULATION AND RESULT
III. ANN MODEL& DATA TRAINING A. Fault Detecting Results
A. ANN Model Construction The ANN model has been well trained so that it can
The complete ANN model is constructed in the detect the fault classification accurately and trip the circuit
MATLAB SIMULINK and shown in figure 3. breaker when a fault occurs. For Single line to ground fault,
the simulation is shown in fig. 4.

Fig. 5. Trip signal for single line to ground (B-G) fault.

From the above figure no 4 it is concluded that at 0.1s the
Fig. 4. Simulink model of the studied transmission line system
transmission line detects a fault signal. At that time the ANN
From figure no 3, it shows that the model contains both the model gives a step input signal immediately and trips the
fault detection and fault classification features. The ANN circuit breaker to isolate the current flow. The ANN trip
networks also present on the Simulink model. signal duration is almost 7ms which means that the ANN
model is highly efficient. When the ANN model detects the
B. Data Training double line to ground fault then the simulation figure is
Utilizing the feed-forward artificial neural network the shown in the below figure no 5.
ANN algorithm is developed through MATLAB/Simulink.
The situation is such that the network is in the state of
readiness to be subjected to being trained and the weights to
be used in the process of training will be selected at random.
When it comes to the training of the artificial neural network,
two methods are followed.
• Supervised training
• Unsupervised training.
In a setting of supervised learning, a training sample
resembling the inputs and outputs of an ideal computation is
introduced to the network, to compute errors such that the
Fig. 6. Trip signal for double line to ground (B-C-G) fault.
actual output of the neural network is closely in line with the
training data. Perceptron, Adaptive linear neuron (Adaline), In Fig. 5. Similarly, it is seen that when the fault occurs at
Backpropagation, generalized delta learning rule are the 0.1s then the ANN model detects the fault by a step input
supervised learning algorithm. The mass under unsupervised signal and trip the circuit breaker isolating the current signal
learning is accommodated with the use of certain in the bus. Here it is also seen that the fault clearing time is
neurobiological principles. These principles can be Hebbian about 8ms. A similar result is observed and shown in figure
learning and competitive Learning process..By taking the 6 for three-phase to ground fault.
superlative number of layers and the probability of validation
and training, the mean square error (MSE) can be lessened in
the training period. In the fault detector neural network, there
are three inputs which are the three-phase currents and one
target and in the fault classifier neural network, there are six
inputs and four target.
C. Data Testing
For the training of a neural network, a huge amount of
data is needed which has been manually processed to set an
output. The errors in this system were sampled with 20000
Fig. 7. Trip signal for three phase to ground (A-B-C-G) fault.
samples/seconds. The fault detector neural n
The fault clearing time is 10ms and it is very fast which is the artificial neural network training. These factor test the
observed in the Fig. 6. There are different factors affect on system as below:
 • Number of layers. ANN model. A MATLAB SIMULINK has been used to
• Percentage of validation and testing. develop the model and the model has been trained by feed
• Number of sampling. forward algorithm. Then it has been observed that the model
In the training period, the initialized data divided into shows satisfactory result to detect and classify the fault.
three data sets, training data, testing data and validation data. Moreover, in this resreach it has been found that the detection
In the final state, if the final error less than the final target and clearing of fault is very fast about (8-10)ms and the error
error, then, the ANN simulates the network. rate of the model is low. In future, the fault location features
can also be added on this proposed protection scheme.

REFERENCES
[1] Painthankar Y. G and Bhide S. R, Fundermentals of Power System
Protection. 2003.
[2] T. Smith, “IEEE Seminar - Fundamentals of Modern Protective
Relaying - Part1,” no. Part 1, 2013, [Online]. Available:
http://ewh.ieee.org/r3/atlanta/ias/2011-2012_Presentations/IEEE
Seminar - Fundamentals of Modern Protective Relaying -
Part1.pdf.
[3] R. P. Nath and V. N. Balaji, “Artificial Intelligence in Power
Systems,” Int. J. Recent Trends Eng. Res., vol. 4, no. 4, pp. 262–
267, 2018, doi: 10.23883/ijrter.2018.4234.eywih.
[4] K. M. Silva, B. A. Souza, and N. S. D. Brito, “Fault detection and
classification in transmission lines based on wavelet transform and
ANN,” IEEE Trans. Power Deliv., vol. 21, no. 4, pp. 2058–2063,
2006, doi: 10.1109/TPWRD.2006.876659.
Fig. 8: Mean-square error performance of the neural network. [5] O. M. Omidvar and D. L. Elliott, Neural Systems for Control.
1997.
[6] P. M. Buscema, G. Massini, M. Breda, W. A. Lodwick, F.
Newman, and M. Asadi-Zeydabadi, “Artificial neural networks,”
Stud. Syst. Decis. Control, vol. 131, no. February, pp. 11–35, 2018,
doi: 10.1007/978-3-319-75049-1_2.
[7] O. I. Abiodun, A. Jantan, A. E. Omolara, K. V. Dada, N. A. E.
Mohamed, and H. Arshad, “State-of-the-art in artificial neural
network applications: A survey,” Heliyon, vol. 4, no. 11, p.
e00938, 2018, doi: 10.1016/j.heliyon.2018.e00938.
[8] B. Ganeshprabu and M. Geethanjali, “Dynamic Monitoring and
Optimization of Fault Diagnosis of Photo Voltaic Solar Power
System Using ANN and Memetic Algorithm,” Circuits Syst., vol.
07, no. 11, pp. 3531–3540, 2016, doi: 10.4236/cs.2016.711300.
[9] D. Howard and B. Mark, “Neural Network Toolbox
Documentation,” Neural Netw. Tool, p. 846, 2004.
[10] A. S. S. Altaie and J. Asumadu, “Fault detection and classification
for compensating network using combination relay and ANN,”
IEEE Int. Conf. Electro Inf. Technol., vol. 2015-June, pp. 351–
356, 2015, doi: 10.1109/EIT.2015.7293367.
[11] A. Prasad and J. B. Edward, “Importance of artificial neural
networks for location of faults in transmission systems: A survey,”
Fig. 9: Linear regression curve of the neural network. Proc. 2017 11th Int. Conf. Intell. Syst. Control. ISCO 2017, pp.
357–362, 2017, doi: 10.1109/ISCO.2017.7856016.
The performance of the neural network can be found from
[12] B. E. M Tayeb, “Faults Detection in Power Systems Using
the mean square error curve fig. 8 and the linear regression
Artificial Neural Network,” Am. J. Eng. Res., vol. 2, no. 6, pp. 69–
curve fig. 9. The performance of the neural network are better
75, 2013, [Online]. Available: www.ajer.org.
because the number of data sets used in the training of
artificial neural network are much enough. [13] A. S. Altaie, “Design of a New Digital Relay for Transmission
Line Fault Detection, Classification and Localization Based on a
V. CONCLUSION New Composite Relay and Artificial Neural Network Approach,”
p. 130, 2015.
In this research the fault detection and classification has
been done successfully in the transmission line through an