International Journal of Research and Engineering
Volume 2, Issue 4
Artificial Neural Network Approach for Discriminating Various faults in Transformer
Protection
1
Ms. A. Pavithra, 2Dr. N. Loganathan
PG Scholar, 2Professor, Department of EEE
K.S.Rangasamy College of Technology, Tiruchengode, India
pavithrapse@gmail.com
1
Abstract — This paper presents a new differential
protection scheme based on Artificial Neural Network
(ANN), which delivers effective distinguish between
internal faults in a power transformer with the other
disturbance such as various types of inrush currents and
overexcitation conditions. In existing method, the
internal faults only considered and the linear
programming method detects the faults at one set value
of the system i.e. it detects the faults only one particular
type and it could not detect all the faults simultaneously.
So the above problem is overcome by detecting the
various types of faults in the system. In the proposed
method Artificial Neural Network (ANN) techniques are
used to detect all the types of faults in the power system.
The Back Propagation Neural Network (BPNN)
algorithm is used to train the process quickly. The
neural network is trained with an input data set and it
gives the output in the following aspects, relay tripping
time and types of faults. The training process for NN and
fault identification result is implemented using toolboxes
on MATLAB/Simulink. Suppose the fault occurs in the
system the ANN will trip the relay and the fault is
isolated from the healthy system. When different types of
faults occurs in the systems are restricted with 0.007 to
0.008s (7 – 8ms). The results endorse that the BPNN is
faster, stable and more reliable to protect the power
transformer from internal faults (three phase to ground
fault, two phase to ground fault and single phase to
ground fault) and other disturbance (overexcitation
condition).
Index
Terms—Power
transformer
protection,
Artificial Neural Network (ANN), Back Propagation
neural network, ANN MATLAB Tool Box.
I. INTRODUCTION
The demand for a reliable supply of electrical energy for
the exigency of the modern world in each and every field
has increased considerably requiring nearly a no-fault
operation of power systems. Power transformers are a class
of very expensive and vital components of electric power
systems. The power transformer is one of the highly
expensive components and it is very important to protect the
transfer from the faults in power system. It is a key
component for electrical energy transfer in a power system.
Stability, reliability and security of the system are important
for the system operation. The maloperation have normally
happened in the relays while using the power transformer
protection. Due to the inrush currents so, there is need of
power transformer protection. Nature of faults in
transformers is internal faults, magnetizing inrush currents
and overexcitation these are normally occurring. The
protections of large transformers are the challenging work to
the protection engineers. The appropriate protection scheme
should need to protect the transformers in order to run the
power system effectively. The relays presently used are
based on differential current principle where filters are
83
ISSN 2348-7852 (Print) | ISSN 2348-7860 (Online)
employed to restrain the second harmonic component and
sometimes the fifth harmonic component in order to avoid
unnecessary tripping against the magnetizing inrush
condition [1], [2]. Kang et al. [4] proposed a transformer
protection technique based on the ratio of the increment in
primary and secondary winding flux linkages.
[3],[7]However, the fundamental limitation of this technique
is that it requires a potential transformer in conjunction with
the current transformer (CT) which further increases the
overall cost of the protection system. Later, Jawad et al. [5]
proposed a decision-making method based on wavelet
transform for discriminating internal faults from the inrush
currents. However, disturbances, such as overexcitation
conditions have not been considered in this paper.
Moreover, Hooshyar et al. [6] presented a method based on
instantaneous frequency for the average differential power
signal to distinguish internal faults from the magnetizing
inrush.
Proposed methods were based on desensitizing or
delaying the relay to overcome the transients [13]. These
methods are unsatisfactory since the transformer may be
exposed for a long unprotected time. Another method based
on the second harmonic content with respect to the
fundamental one was introduced, known as harmonic
restraint differential protection [14], which improved
security and dependability was appreciated. It has been
observed by the authors that none of the researchers have
considered special types of internal faults, such as turn-toturn and primary-to-secondary winding. Artificial Neural
Networks (ANN) are extremely used, particularly in the
field of power system protection since 1994. [8],[9]The
main advantage of the ANN method over the conventional
method is the non-algorithmic parallel distributed
architecture for information processing and inherent ability
to take intelligent decision. In recent years, few works which
investigate the feasibility of using ANN for power
transformer differential protection has also been reported
[15–21]. The feasibility of the proposed scheme has been
tested over a test data set of 4 cases of modeling an existing
three-phase power transformer of MATLAB/Simulink
software package. A comparison of the proposed scheme
with the conventional harmonic restrains scheme indicates
the superiority of the proposed scheme providing an overall
discrimination accuracy of more than 90%.
II. POWER TRANSFORMER PROTECTION
Figure 1 shows the model of 3ɸ, 50 Hz, 120 kV/25 kV,
47 MVA. The modeling is performed using the software
package. Three-phase differential current samples for onecycle duration are acquired through CTs connected on both
sides of the power transformer. Nonlinearity due to CT
saturation and phase compensation conditions is also
considered for generating the simulation cases. The method
used for generating various simulation cases for different
types of internal faults and other disturbances is explained in
the subsequent sections.
http://www.ijre.org
International Journal of Research and Engineering
120/25 kV, 47 MVA
Fig. 1.
Volume 2, Issue 4
adjusting the node weights and biases accordingly. The
speed of processing, allowing real time applications, is also
an advantage. Figure 2. Shows the typical three-layer
architecture of an ANN.
Block Diagram of Power Transformer
Protection
During power transformer operation, it encounters any
one of the following conditions:

Internal faults

Other Disturbance
A. Internal faults
Internal faults consider the single phase to ground fault,
two phases to ground fault, three phase to ground fault and
other disturbance considers the over excitation condition.
This paper has been simulated in MATLAB/Simulink
software using single phase to ground fault, two phases to
ground fault, three phase to ground fault.
B. Other Disturbance
In order to avoid tripping of the differential protection
scheme during an overexcitation condition, a separate
transformer overexcitation circuit should be used [8]. In
order to check this phenomenon, various overexcitation
conditions are simulated with different values of terminal
voltage varying from 105% to 125% of rated voltage of the
power transformer in steps of 5% with 5% variation in
fundamental frequency. The simulation result shows that the
proposed algorithm helps in the protection of power
transformer and to distinguish the internal fault and other
disturbance.
Typical three-layer architecture of an
ANN
IV. artificial Neural Network training process
A. Back Propagation Neural Network Algorithm
The algorithm is explained from the flowchart of back
Propagation network as shown in the figure 3. The
algorithm steps as follows:
Step 1: Start the neural network training set
Step 2: Define the train data
Step 3: When the program is run, the tool box wopen
Step 4: Fix the parameter values (epoch, gradient,
learning rate) in ANN toolbox
Step 5: Start the training process
Step 6: Check the goal level
Step 7: If yes compute the weight for each of the
preceding layer by back propagation error if
no it will repeat the training process
Step 8: Stop the process
Fig. 2.
III. ARTIFICIAL NEURAL NETWORK BASED
POWER TRANSFORMER PROTECTION
SCHEME
The ANNs do not need a knowledge base to work.
Instead, they have to be trained with numerous actual cases.
An ANN is a set of elementary neurons, which are
connected together different architectures organized in
layers what is biologically inspired shown in figure 4. An
elementary neuron can be seen like a processor which makes
a simple nonlinear operation of its inputs producing its
single output. A weight (synapse) is attached to each neuron
and the training enables adjusting of different weights
according to the training set. The ANN techniques are
attractive because they do not require tedious knowledge
acquisition, representation and writing stages and, therefore,
can be successfully applied for tasks not fully described in
advance. The ANN is not programmed or supported by a
knowledge base as being Expert Systems. Instead, they learn
a response based on giving inputs and a required output by
84
ISSN 2348-7852 (Print) | ISSN 2348-7860 (Online)
http://www.ijre.org
International Journal of Research and Engineering
Volume 2, Issue 4
derivatives are being calculated. Figure 4. Shows the
artificial neural network training toolbox.
Fig 4. Artificial Neural Network Training Toolbox
C. Neural Network Training performance
Fig 3. Back Propagation Network
B. Neural Network Training Toolbox
Neural Network Toolbox™ provides functions and apps
for modeling complex nonlinear systems that are not easily
modeled with a closed-form equation. Neural Network
Toolbox supports supervised learning with feedforward
networks. It also supports unsupervised learning with selforganizing maps and competitive layers. With the toolbox
can design, train, visualize, and simulate neural networks.
Use the Neural Network Toolbox for applications such as
data fitting, pattern recognition, clustering, time-series
prediction, and dynamic system modeling and control.They
are three algorithms are used in neural network toolbox,
 Levenberg-Marquardt back propagation
 Mean squared normalized error performance function
 Default derivative
Levenberg-Marquardt algorithm is specifically designed
to minimize the sum-of-square error functions. Mean
squared normalized error performance function of
an estimator measures the average of the squares of the
"errors", that is, the difference between the estimator. MSE
is a risk function, corresponding to the expected value of the
squared error loss or quadratic loss. The difference occurs
because of randomness or because the estimator doesn't
account for information that could produce a more accurate
estimate. The MSE is the second moment (about the origin)
of the error, and thus incorporates both the variance of the
estimator and its bias. For an unbiased estimator, the MSE is
the variance of the estimator. Like the variance, MSE has
the same units of measurement as the square of the quantity
being estimated. This function chooses the recommended
derivative algorithm for the type of network whose
85
ISSN 2348-7852 (Print) | ISSN 2348-7860 (Online)
Plotperform(TR) plots the training, Best and goal given
the training record TR returned by the function train.Figure
5. shows the neural network training performance.The best
training performance is 1.4672e-11 at epoch 25.The best
training performance of the goal is met in the Back
propagation neural network.
Fig 5. Neural Network Training Performance
D. Neural Network Training Regression
Plotregression (targets, outputs) plots the linear
regression of targets relative to output. Figure 6 shows the
neural network training regression.
http://www.ijre.org
International Journal of Research and Engineering
Fig 6. Neural Network Training Regression
2. SIMULATION RESULTS
According to the performance and training regression
process is taken for the below faults.
A. Three Phases To Ground Fault
Figure 7. Shows the simulation results from three phase to
ground fault. The three phase to ground fault occurs at the
instant t = 0.074s and fault tripping time is 0.0072s
Fig 7. Three Phases to Ground Fault and its
Tripping Time
B. Two Phase to Ground Fault (Phase B & C)
Figure 8. Shows the simulation results for two phases (B
& C) to ground fault. The two phases (B &C) to ground
fault occurs at the instant t = 0.202s and fault tripping time
is 0.008s
Fig 8. Two Phase to Ground Fault (Phase B & C)
C. Single Phase to Ground Fault (Phase A)
Figure 9. Shows the simulation results for Single Phase
to Ground Fault (Phase A). The Single Phase to Ground
Fault (Phase A) occur at the instant t = 0.412s and fault
tripping time is 0.008s
Fig 9. Single Phase to Ground Fault (Phase A)
D. Overexcitation Condition
Figure 10. Shows the simulation results for overexcitation
condition. The overexcitation condition occurs at the instant
t = 0.518s and fault tripping time is 0.0073s.
86
ISSN 2348-7852 (Print) | ISSN 2348-7860 (Online)
Volume 2, Issue 4
Fig 10. Overexcitation Condition
All the above faults are tripped with in the very short
period when compared to conventional relay fault tripping
time, this because of only the ANN technique used.
V. CONCLUSION
A new approach for the differential protection scheme
based on ANN is effectively distinguished the internal faults
and other disturbances in a power transformer. The input of
the dataset is given to the Neural Network tool box for fault
classification for tripping the fault current. The NN is
trained with the features extracted for the different fault
conditions. The proposed ANN techniques are showed their
fast response for tripping the fault and it trip the fault with in
0.007 to 0.008s. The proposed algorithm provides more
accurate results. Distinguishing the internal faults and other
disturbances cases have been generated in the
MATLAB/Simulink software package. The proposed
algorithm has considered the four test cases which include
three phases to ground fault, two phases to ground fault,
phase to ground fault and over excitation conditions. In all
cases the ANN discriminates the internal faults and other
disturbances. The tripping time is reduced and gives the
quick response.
References
[26] P. M. Anderson, Power System Protection. New York:
IEEE, 1999.
[27] Y. V. V. S. Murty and W. J. Smolinski, ―A Kalman
filter based digital percentage differential and ground
fault relay for a 3-phase power transformer,‖ IEEE
Trans. Power Del., vol. 5, no. 3, pp. 1299–1308, Jul.
1988.
[28] M.-C. Shin, C.-W. Park and J.-H. Kim, ―Fuzzy logicbased relaying for large power transformer protection,‖
IEEE Trans. Power Del., vol.18, no. 3, pp. 718–724,
Jul. 2003.
[29] Y. C. Kang, B. E. Lee, S. H. Kang., and P. A. Crossley,
―Transformer protection based on the increment of flux
linkages,‖ Proc. Inst. Elect. Eng., Gen. Transm. Distrib.,
vol. 151, no. 4, pp. 548–554, Jul. 2004.
[30] J. Faiz and S. Lotfi-Fard, ―A novel wavelet-based
algorithm for discrimination of internal faults from
magnetizing inrush currents in power transformers,‖
IEEE Trans. Power Del., vol. 21, no. 4, pp.1989–1996,
Oct. 2006.
[31] A. Hooshyar, S. Afsharnia, M. Sanaye-Pasand, and B.
M. Ebrahimi, ―A new algorithm to identify magnetizing
inrush conditions based on instantaneous frequency of
differential power signal,‖ IEEE Trans. Power Del., vol.
25, no. 4, pp. 2223–2233, Oct. 2010.
[32] S. H. Horowitz and A. G. Phadke, Power System
Relaying, 3rd New York: Wiley, 2008, pp. 201–205.
[33] T. S. M. Rao, Power System Protection-Static Relays.
New Delhi, India: McGraw-Hill, 1979.
http://www.ijre.org
International Journal of Research and Engineering
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
N. Cristianini and J. Shawe-Taylor, An Introduction to
Support Vector Machines and Other Kernel-Based
Learning Methods, 1st ed. Cambridge, U.K.:
Cambridge Univ. Press, 2000.
C. C. Chang, LIBSVM-A library for support vector
machines, 2011
S. V. Kulkarni and S. A. Khaparde, Transformer
Engineering Design and Practice. New York: Marcel
Dekker, 2005.
P. Bastard, P. Bertrand, and M. Meunier, ―A
transformer model for winding fault studies,‖ IEEE
Trans. Power Del., vol. 9, no. 2, pp.690–699, Apr.
1994.
P. Arboleya, G. Diaz, J.G. Aleixandre, ―A solution to
the dilemma inrush/fault in transformer relaying using
MRA and wavelets‖, Electric Power Compo. Syst. 34
(3), pp.285–301, 2006.
S.A. Saleh, M.A. Rahman, ―Modeling and protection of
a
three-phase transformer using wavelet packet
transform‖, IEEE Transactions on Power Delivery. 20
(2), pp. 1273–1282, 2005.
L. Yongli, H. Jiali, D. Yuqian, ―Application of neural
network to microprocessor based transformer protective
relaying‖, IEEE Int. Conf. on Energy Management and
Power Delivery, vol. 2, 21–23, pp. 680–683, November
1995.
L.G. Perez, A.J. Flechsing, J.L. Meador, Z. Obradovic,
―Training an artificial neural network to discriminate
between magnetizing inrush and internal faults‖, IEEE
Transactions on Power Delivery. 9 (1), 434–441, 1994.
P. Bastard, M. Meunier, H. Regal, ―Neural network
based algorithm for power transformer differential
relays‖, IEE Proceedings Generation. Transmission and
Distribution. 142 (4), 386–392, 1995.
J. Pihler, B. Grcar, D. Dolinar, ―Improved operation of
power transformer protection using artificial neural
network‖, IEEE Transactions on Power Delivery. 12
(3),1128 – 1136, 1997.
M.R. Zaman, M.A. Rahman, ―Experimental testing of
the artificial neural network based protection of power
87
ISSN 2348-7852 (Print) | ISSN 2348-7860 (Online)
Volume 2, Issue 4
[45]
[46]
transformers‖, IEEE Transactions on Power Delivery.
13 (2), 510–517, 1998.
.L.Orille-Fernandez,N.K.I.Ghonaim, J.A.Valencia, ―A
FIRANN as differential relay for three phase power
transformer protection‖, IEEE Transactions on Power
Delivery.16 (2),215-218, 2001.
H. Khorashadi-Zadeh, ―Power transformer differential
protection scheme based on symmetrical component
and artificial neural network‖, IEEE Int. Conf. on
Neural Network Application in Electrical Engineering,
23–25, pp. 680–683, September, 2004.
http://www.ijre.org