Content uploaded by Anamika Yadav
Author content
All content in this area was uploaded by Anamika Yadav on Jun 23, 2016
Content may be subject to copyright.
1
FAULT DETECTION, CLASSIFICATION AND SECTION
IDENTIFICATION ON DISTRIBUTION NETWORK WITH D-STATCOM
USING ANN
Garima Netam1*and Anamika Yadav2
Department of Electrical Engineering, National Institute of Technology,
Raipur, CG, India1,2
Received: 14-May-2016 / Accepted: 20-May-2016
©2016 ACCENTS
Abstract
This paper presents easy and efficient method of
fault detection, fault classification and section
identification in distribution networks with
Distribution Static Var Compensator (D-
STATCOM) using Artificial Neural Networks. The
neural network uses Levenberg-Marquardt
Backpropagation algorithm for training. The D-
STATCOM (average) model available in MATLAB
has been modified to perform fault analysis. D-
STATCOM is used for reactive power
compensation, and regulates the system voltage by
absorbing and generating reactive power. Fault is
simulated for different function of D-STATCOM in
which it absorbs reactive power like an inductor and
generate reactive power like a capacitor. The
present work reports the results of fault detection,
fault classification and section identification
whether it is forward fault and reverse fault in
distribution network with D-STATCOM.
Keywords
Distributed network, D-STATCOM, ANN, fault detection,
section identification, fault classification, Levenberg-
Marquardt Backpropagation algorithm
1. Introduction
Nowadays, the Flexible alternating current
transmission system (FACTS) device is widely used
in transmission and distribution system to enhance
power transfer capability and stability of the power.
There are so many devices which are used in
transmission line but D-STATCOM is a shunt device
which is used only in distribution networks. D-
STATCOM is similar to STATCOM, but only
difference is STATCOM is used in transmission
level. The performance of D-STATCOM system is to
reduce power quality problems and improve
distribution system performance under all types of
system related disturbances and system unbalance
faults, such as Line-to-Line (LL) and Double Line to
Ground (DLG) faults etc [16]. It improve the power
quality like voltage fluctuations, voltage sag, voltage
swell etc. The D-STATCOM operates when there is a
sudden change in source voltage (increase/decrease),
it absorbs or generates reactive power. In D-
STATCOM two Voltage Source Converter (VSC)
and capacitor is used to provide the dc voltage to the
VSC.
The two modelling approaches (detailed and
average model) are explained by the researcher in a
+3/-3 Mvar D-STATCOM on a 25kV network by
comparing their static and dynamic performance. The
detailed and average model is compared when system
is changes from inductive to capacitive operation.
Chopping of dc voltage is observe on detailed where
as average value is observe in average model [1].
Several method exist for detecting , classifying and
section identification on transmission line[2-15].
ANN is widely used for fault detection, fault
classification and section identification. The
researcher worked on fault detection, fault
classification and direction estimation for
uncompensated transmission lines [2-11]. Directional
relay based on negative or zero sequence components
or compensated post fault voltages are almost
common [2, 3] but the relay is slowly operating for
all types of faults, travelling waves is proposed for
reducing the operating time of directional relay [4, 5].
The protection over doubly fed transmission line [2,
6] greater portion of the line length can be protected
comparing to earlier techniques. The input training
data of the recording devices was sampled using
digital signal processing [7] for fault classification.
The protective relay scheme based on discrete
wavelet transform and ANN provide primary and
backup protection of 99% to the line within quarter
cycle time [8]. The adaptive neuro-fuzzy inference
system approach is used for directional estimation
insectional transmission lines [9]. Fuzzy inference
2
system is used to improve the performance of
directional relaying, fault classification in
transmission line. The fuzzy logic based relay is less
complex and better than other AI based methods such
as ANN, support vector machine, SVM and decision
tree because training patterns is not required [10].
The researcher worked on fault detection, fault
classification and direction estimation for
compensated transmission lines [12-15].
Superimposed sequence components-based integrated
impedance (SSCII) technique is used for fault
detection and fault classification in shunt ( static var
compensator ) compensated line. The magnitude of
SSCII is small for internal fault and very large for
external faults [12]. ANN and wavelet transform is
used for the protection of transmission line with
FACTS ( Thyristor-Controlled Series Capacitor )
device [13]. Fault detection and direction estimation
based on ANN [14]. The fault detection and fault
classification is based on Wavelet Entropy used in
FACTS Compensated Transmission lines, the
FACTS device are static synchronous series
compensators (SSSC) and Unified Power Flow
Controller (UPFC) which is placed in the midpoint of
the transmission line [15].
This paper presents Artificial Neural Network
algorithm is applied for the fault detection, fault
classification and section identification for forward
and reverse fault in distribution network with D-
STATCOM. Various combination of fault resistance,
fault inception angle, fault distance and type of fault
are considered. The data used in this technique is
obtained from the simulate model of D-STATCOM
on a 25kV network, which is done on MATLAB. The
result based on this algorithm is simple, fast and
accurate.
2. Operation of D-STATCOM
D-STATCOM is a shunt device which is used to
regulate the system voltage by generating and
absorbing reactive power. The network is connected
to the D-STATCOM through the transformer and D-
STATCOM consists of PWM inverter and PWM
inverter consist of two IGBT bridges. At the DC side
of inverter, a capacitor provides dc link voltage and
that capacitor takes power from the network for
charging. The controller of D-STATCOM provides
control to dc link voltage and bus voltage. The main
role of D-STATCOM is to synchronize the bus
voltage by generating and absorbing reactive power
just like a TSC (thyristor static compensator). The
transfer of reactive power between the network and
D-STATCOM is possible through the leakage
reactance of the coupling transformer by using a
secondary voltage in phase with the primary voltage
(network side). Secondary side is D-STATCOM and
primary side is network. There are two conditions for
operation which are:
1. If the bus voltage is higher than the
secondary voltage then the D-STATCOM
absorbs reactive power like an inductor.
2. If the bus voltage is lower than secondary
voltage then the D-STATCOM generates
reactive power like a capacitor.
V1
V2
Q
D-STATCOM BUS
XL
I
(a)
V1
V2
Q
D-STATCOM BUS
XL
I
(b)
Figure 1: D-STATCOM operation (a) inductive
operation (b) capacitive operation
In steady state condition D-STATCOM produces
small active power to compensate the inverter losses
and the bus voltage leads the inverter voltage by a
small angle.
3. Modelling of D-STATCOM
The D-STATCOM is used to regulate the parameters
of 25kV distributed network. We consider here, a
+3/-3Mvar D-STATCOM connected to distribution
network [1]. Two feeders of 21 km and 2 km are
there which transmit power to the load. Feeder of 21
km is modeled by pi-equivalent circuit and connected
between the bus 1 and bus 2, whereas 2 km feeder is
connected between bus 2 and bus 3. For power factor
correction, a shunt capacitor is added at bus 2. Two
different loads are connected to the bus 4 through
distributed transformer of 25kV/600V and which are
3
resistive load of 1MW and variable load. In D-
STATCOM two voltage source converters are used
which is consist of IGBT, 10,00 microfarad capacitor
is used to provide dc to the voltage source converter
and we get ac from the output side. An LC filter is
used in output side of the inverter to provide quality
factor and a resistance is connected in series with the
capacitor. Voltage and current acquisition is done
using anti-aliasing filter. Controller of D-STATCOM
consists of Phase Locked Loop, and two
measurements system (voltage and current) and DC
voltage regulator.
Figure 2: Simulink model of Distribution Network
with D-STATCOM
3.1 Measurement of voltage and current and
preprocessing of data
At sampling frequency of 1,200 Hertz, 20 samples
per each cycle, voltage and current waveforms have
been generated 0 50 100 150 200 250 -3 -2 -10123 x
104 No of samples Amplitude of voltage at bus 2
(volts) 0 50 100 150 200 250 -1500 -1000 -5000 500
1000 1500 No of samples Amplitude of current at bus
2 (amps) from the model. The starting of one cycle
obtained high value of voltage and current waveform
because of the transient. Transient occurs due to
switching operation of devices. The fault occur after
the voltage source is increased by 6% at the time of
20 ms. The voltage is decreases and current is
increases during the fault. The Power system model
is simulated in the MATLAB Simulink software and
obtained fundamental components of three phase
voltage and current by using Discrete Fourier
transform (DFT) from bus 1,2 and 3. The
fundamental components of three phase voltage and
current of bus 2 is used as an input to the ANN
module. The figures 3 shows voltage and current
waveforms of phase A-G (ground) fault in forward
section at 1-km from Bus-2, 10-ohm fault resistance
and 0-degree fault inception angle. The training
process is fast and improved by using preprocessing
of three phase voltage and current signals.
050 100 150 200 250
-3
-2
-1
0
1
2
3x 104
No of samples
Amplitude of voltage at bus 2 (volts)
(a)
050 100 150 200 250
-1500
-1000
-500
0
500
1000
1500
No of samples
Amplitude of current at bus 2 (amps)
(b)
Figure 3: (a) Voltage and (b) Current waveforms
of bus 2 during phase C-G (ground fault)
4. Artificial Neural Network
The artificial neural network (ANN) is an
information processing model inspired by the
biological nervous system. An ANN is composed of a
large number of highly interconnection processing
elements (neurons) working in unison to solve
specific problems. An ANN is configured for a
specific application such as pattern recognition or
data classification through learning process [17].
Artificial neural network is fast, simple, efficient and
accurate technique. The training patterns is required.
It is used for linear as well as non-linear model.
ANN is a programming technique, and can easily
solve the linear and non-linear problems for fault
detection, fault classification and section
identification. For best results, the factors that need to
be appropriately selected are
4
Figure 4 shows the architecture of the neural network
Figure 4: Architecture of ANN
5. Training and Testing
The model of distribution network with D-
STATCOM is built in MATLAB/ Simulink and ANN
algorithm is used. The Levenberg-Marquardt
backpropagation based algorithm is used for training
the neural network. The training data set consists of
total (3 fault resistances * 39 fault lengths * 3 fault
inception angles * 11 types of faults = 3861 ) fault
cases and one no-fault case is added. Since, it is a
supervised learning method so Target is needed for
learning process. Training is done by taking
activation function which are (tansig, tansig, purelin)
and three hidden layers (20, 20,1) and checking the
regression plot is accurate.
Table 1: Trainning Pattern
Training pattern
Fault length (km)
Reverse section is 21 km, fault
increase in each 1 step
forward section is 2 km, fault increase
in each 0.1 step
Fault inception
angle in degree
0 , 45 , 90
Fault resistance
(ohm)
0, 50, 100
Type of fault
A-G, B-G, C-G, AB, BC, CA, AB-G, BC-G,
CA-G, ABC, ABC-G
5.1 Fault Detection
For fault detection, six inputs are provided to the
neural network, which are three phase voltages (VA,
VB and VC) and three phase currents (IA, IB and
IC). For developing of data set for training, all
different eleven types of faults and no-fault
conditions are also considered. The training data set
consists of total (3 fault resistances * 39 fault lengths
* 3 fault inception angles * 11 types of faults = 3861
) fault cases and one no-fault case is added. The
output of neural network is in form of „0‟ and „1‟
(„no‟or „yes‟), which shows „0‟for nofault data and
„1‟ for faulted data. Figure 5(a)shows the testing
result at 5.45 km reverse section from bus 2, 30 ohm
fault resistance, 30 degree fault inception angle and
phase B-C fault. Figure 5(b) shows the testing result
at 0.55 km forward section from bus 2, 10 ohm fault
resistance 30 degree fault inception angle and phase
BC-G (ground) fault.
5.2 Section Identification
The six inputs provide to the neural network and
getting 2 outputs. The section identification is done at
middle bus (bus2) and it check whether the fault has
occurred in forward direction or reverse direction.
Training is done with three activation function‟s
(tangig, tansig, purelin), three hidden layer (20, 20, 2)
and checking the regression plot is accurate as same
process employed in detection. After testing the two
output get from neural network in the form of „0‟ or
„1‟. First output is reverse section and second is
forward section. Figure 6(a) shows the testing result
at 5.45 km reverse section from bus 2, 30 ohm fault
resistance, 30 degree fault inception angle and phase
B-C fault Figure 6(b) shows the testing result at 0.55
km forward section from bus 2, 10 ohm fault
resistance, 30 degree fault inception angle and phase
BC-G (ground) fault.
5.3 Fault Classification
The same process which was employed in detection
is used in classification. Six inputs ( three phase
voltage and three phase current) are taken but the
neural network takes four outputs, each of them is
three phase for each fault condition and one is for
ground fault. The output is in form of „1‟ or „0‟ („yes‟
or „no‟). According to various fault condition, various
possibilities can be represented. The neural network
which is proposed can distinguish accurately between
eleven faults.
For classification, two training sets have been
taken because training matrix is very large, the
complexity increases and it takes more time to train
the data. One data set contains single line to ground
fault, double line to ground fault and triple line to
ground fault. Second data set contains double line
fault and triple line fault. Training is done using three
hidden layers ( 20, 20, 4 ) and activation functions
5
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2
-4
-2
0
2
4x 104
time (sec)
Three phase Voltag e (volts)
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2
-1000
-500
0
500
1000
time (sec)
Three phase Cur rent (amps)
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2
-1.5
-1
-0.5
0
0.5
1
1.5
time (sec)
outpu t
(a)
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2
-4
-2
0
2
4x 104
time (sec)
Three phase Voltag e (volts)
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2
-2000
-1000
0
1000
2000
time (sec)
Three phase Cur rent (amps)
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2
-0.5
0
0.5
1
1.5
time (sec)
outpu t
(b)
Figure 5: (a). Three phase Voltage, Current and
detector output during phase B-C fault at 5.45 km
from Bus 2 with fault resistance of 30 ohm at
time51.3891ms.
(b). Three phase Voltage, Current and detector
output during phase B-C fault at 0.55 km from
Bus 2 with fault resistance of 10 ohm at
time51.3891ms.
0 0.05 0.1 0.15 0.2
-1
0
1
time (sec)
reverse section outpu t
0 0.05 0.1 0.15 0.2
-1
0
1
time (sec)
forwar d section outpu t
(a)
0 0.05 0.1 0.15 0.2
-1
0
1
time (sec)
reverse section outpu t
0 0.05 0.1 0.15 0.2
-1
0
1
time (sec)
forwar d section outpu t
(b)
Figure 6: (a). Reverse section output and Forward
section output during phase B-C fault at 5.45 km
from Bus 2 with fault resistance of 30 ohm at
time51.3891ms.
(b). Reverse section output and Forward section
output during phase B-C fault at 0.55 km from
Bus 2 with fault resistance of 10 ohm at
time51.3891ms.
(tansig, tansig, purelin). The regression plot and
performance curve prove that the training is accurate.
After training, testing is done by taking single fault
resistance, single fault inception angle and single
fault length with one type of fault. The test results for
fault classification is shown in figure in which taken
same parameters as used in fault detection. In fault
classification the relay operates for forward direction,
the results shows for forward section only. Figure 7
shows the testing result at 0.55 km forward section
from bus 2, 10 ohm fault resistance, 30 degree fault
inception angle and phase BC-G (ground) fault.
6
0 0.05 0.1 0.15 0.2
-1
0
1
outp ut A
0 0.05 0.1 0.15 0.2
-1
0
1
outp ut B
0 0.05 0.1 0.15 0.2
-1
0
1
outp ut C
0 0.05 0.1 0.15 0.2
-1
0
1
time(sec)
outp ut G
Figure 7: Test result for BC-G fault at 0.55 km
from Bus 2 at fault inception time is 51.3891ms
and fault resistance of 10 ohm
6 Conclusion and Future Scope
The proposed method is very simple and accurate
technique for fault detection, fault classification and
section identification (forward or reverse) which
gives successful result under various fault conditions.
The fundamental components of three phase voltage
and current is used as an input to the neural network.
For training Levenberg-Marquardt, Backpropagation
algorithm is used in neural network. Advantage of the
proposed method is suitable for different fault
conditions such as fault resistance, fault inception
angle, fault type, fault distance, etc. The result gives
high accuracy and exact operation of proposed fault
detection, classification, section identification
approach.
With advent of smart grid technologies, the use of
ANN based relays will help in fast fault analysis
replacing the conventional relays and the use of D-
STATCOM will help to maintain the voltage profile
thereby increasing the reliability over distribution
network.
Appendix
System Parameters of Figure 2 (Base MVA=100)
A 3-phase ac source:
Rated voltage: 25kV
Line parameter of 21km Feeder:
Positive sequence resistance: 0.1153 Ω/km
Zero sequence resistance: 0.3963Ω/km
Positive sequence inductance: 1.048e-3 H/km
Zero sequence inductance: 2.730e-3 H/km
Positive sequence capacitance: 11.33e-9 F/km
Zero sequence capacitance: 5.338e-9 F/km
Line length: 21km
Line parameter of 2km Feeder:
Positive sequence resistance: 0.1153 Ω/km
Zero sequence resistance: 0.3963Ω/km
Positive sequence inductance: 1.048e-3 H/km
Zero sequence inductance: 2.730e-3 H/km
Positive sequence capacitance: 0 F/km
Zero sequence capacitance: 0 F/km
Line length: 2km
25kV +/-3 Mvar DSTATCOM:
DC voltage (controller): 2400
Cut off frequency: 2000
Capacitor: 10000e-6 F
Transformer:
Rated voltage: 25kV/600V
Load:
Load1: 1MW
Load2: variable load
Nominal voltage: 600V
Current: 3000 amp
Sample time: 4e-5 s
Acknowledgment
The author wish to acknowledge the Department of
Electrical Engineering, National Institute of
Technology Raipur For Research facilities provided
to conduct this project.
7
References
[1] Maqsud Hesenzade Mehemmed Oglu, Eliyev
Elovser Qaraca Oglu, “Modelling and
Simulation of a Distribution STATCOM using
simulink‟s power system Blockset,”: The 5th
International Conference on Application of
Information and Communication Technologies
(AICT), 2011, pp. 1-5.
[2] Anamika Jain and A.S. Thoke “, “ANN Based
Directional Fault Detector/Classifier for
Protection of Transmission Lines,” International
Journal of Computer Science & Information
Technologies, (IJCSIT), Tamil Nadu, India, Vol
2 (5) 2011, 2426-2433.
[3] J.L. He, Y.H. Zhang and N. C. Yang, “New type
power line carrier relaying system with
directional comparison for EHV transmission
lines,” IEEE Transactions on Power Apparatus
and system, vol. PAS-103, no. 2, pp. 429-436,
February 1984.
[4] A.T. Johns, “New ultra-high-speed directional
comparison technique for the protection of EHV
transmission lines,” IEE PROC, vol. 127, no. 4,
JULY 1980.
[5] Y.Q. Xia, J.L. He and K.K. Li, “A reliable
directional relay based on compensated voltage
comparison for EHV transmission lines,” IEEE
Transactions on Power Delivery, vol. 7, no. 4,
October 1992 , pp. 1955-1962.
[6] Anamika Yadav and A.S. Thoke, “Transmission
line fault distance and direction estimation using
artificial neural network”, International Journal of
Engineering, Science and Technology, Vol. 3,
No. 8, 2011, pp. 110-121.
[7] Ahmed Sabri Salman Altaie, J. Asumadu, “fault
detection and classification for compensating
network using combination relay and ANN” 2015
IEEE international Conference on
Electro/Information Technology (EIT), 21-23
May 2015, pp.351-356.
[8] Aleena Swetapadma and Anamika Yadav, “Time
Domain Complete Protection Scheme for Parallel
Transmission Lines”, Ain Shams Engineering
Journal, Dec 2015 Elsevier, Impact Factor-0.932.
[9] Aleena Swetapadma and Anamika yadav, “High
Speed Directional Relaying Using Adaptive
Neuro-Fuzzy Inference System and Fundamental
Component of Currents”, IEEJ Transactions on
Electrical and Electronic Engineering, Volume
10, Issue 6, pages.
[10] Anamika Yadav , Aleena Swetapadma,
“enhancing the performance of transmission line
directional relaying, fault classification and
fault location schemes using fuzzy inference
system” IET Generation, Transmission &
Vol.9,Iss. 6, October 2014, pp. 580-591, Impact
Factor-1.414.
[11] M. Jamil, S.K. Sharma and R. Singh, “fault
detection and classification in electrical power
transmission system using ANN” springerplus,
v.4, 2015PMC4496419.
[12] M. Jamil, S.K. Sharma and R. Singh, “fault
detection and classification in electrical power
transmission system using ANN” springerplus,
v.4, 2015PMC4496419.
[13] A. Y. Abdelaziz, Amr M. Ibrahim, “A Hybrid
Wavelet-ANN_Based Protection Scheme for
FACTS Compensated Transmission lines,” IJ
Intelligence Systems and Applications, 04 June
2013, pp 23-31.
[14] Aditi Verma and Anamika Yadav, "ANN Based
Fault Detection & Direction Estimation Scheme
for Series Compensated Transmission Lines"
Paper ID EE 9037, 2015 IEEE International
Conference on Electrical, Computer and
Communication Technologies (IEEE ICECCT
2015), 05 - 07, March 2015, SVS College of
Engineering, Coimbatore, Tamil Nadu, India.
[15] Amany M. EI-Zonkoly, Hussein Desouki,
“Wavelet Entropy Based Algorithm for fault
Detection and Classification in FACTS
Compensated Transmission line” Energy and
Power Engineering, 2011, 3 , pp 34-42.
[16] R. Lakavath, K. Kumaraswamy, S.Sahithi and G.
Naresh, “Design and implementation of Flexible
D-Statcom for Mitigation Power Quality
Problems and Improve the Distribution System
Performance,” Int. Journal of Engineering
Research and Applications, ISSN: 2248-9622,
Vol. 4, Issue 1(Version 2), January 2014, pp.404-
413.
[17] S.N. Sivanandam and S.N. Deepa, Principles of
Soft Computing, copyright © 2011 by Wiley
India Pvt. Ltd., 4435-36/7, Ansari Road,
Daryaganj, New Delhi 110002.
Garima Netam graduated B.E. in
Electrical Engineering from GIMT,
Chhattisgarh Swami Vivekanand
Technical University, CG, India, in the
year 2013 and pursuing her M.Tech
from National Institute of Technology,
Raipur, CG, India.
Anamika Yadav did her B.E. in
Electrical Engineering from RGPV
Bhopal, India, in the year 2002. She
acquired her M.Tech Degree in
Integrated Power System from V.N.I.T.,
Nagpur, India, in 2006 and Ph.D.
degree in Electrical Engineering from
CSVTU, Bhilai, with research centre at
NIT, Raipur, C.G., India, in 2010. She worked as Assistant
Engineer in the Chhattisgarh State Electricity Board,
Raipur, CG, India, for 4.8 years. She joined National
Institute of Technology Raipur, CG, India, on March 2009
as Assistant Professor in the Department of Electrical
8
Engineering. Her research interest includes application of
soft computing techniques to Power System protection.
Recently she has been elevated to senior member IEEE.
She is also the member of IET, IE(I) since 2009.
