Journal of Multidisciplinary Engineering Science and Technology (JMEST)

ISSN: 2458-9403
Vol. 7 Issue 6, June - 2020

Fault Detection, Classification and

Location on 132kv Transmission Line based
on DWT and ANFIS
Francis Onette OGBAN1 Kingsley Monday UDOFIA2 Chidinma NnekwuKALU3
Department of Department of Department of
Electrical/Electronic and Electrical/Electronic and Electrical/Electronic and
Computer Engineering, Computer Engineering, Computer Engineering,
University of Uyo, Akwa Ibom, University of Uyo, Akwa Ibom, University of Uyo, Akwa Ibom,
Nigeria Nigeria Nigeria
[email protected] Corresponding Author: [email protected]
[email protected]

Abstract— In this paper, a robust fault detection, systems. Power system components are always
classification and location scheme for 132 kV, 100 km subjected to the greatest stresses from excessive
transmission line is developed. This scheme combines currents as a result of fault [2].
the feature extraction capability of the discrete wavelet Faults on transmission line require fast restoration
transform and the intelligent classification capability of because it generates damage in the power system.
adaptive neuro-fuzzy inference system (ANFIS). RMS So an accurate, fast and reliable method to detect,
values gotten from the five detail levels (Daubechies classify and locate the fault is needed to be
dB5 wavelet) of the DWT decomposition of the current established on the transmission lines to insure healthy
signals were used for the training of the ANFIS models. power system [3]. Over the years, several fault
Simulation investigations were done for different fault detection and classification methods on transmission
scenarios with faults at different phases, locations, fault lines have been used. They are classified into
resistance and fault inception angles. From the analysis conventional and intelligent methods [4]. Conventional
of the results obtained, it was obvious that the scheme method of fault detection and classification is based
was able to discriminate between actual fault cases on the variation of voltage and current lines. In
from normal condition in a maximum time of 8 conventional methods phase current or voltage or
milliseconds after fault inception. The classification of instantaneous values of current or voltage are directly
fault types was exact and the locations of the faults used as relay inputs [5]. The intelligent method is
were identified with a maximum RMSE of 0.109 km and based on fuzzy logic, artificial intelligence, support
a maximum absolute error of 0.22 km which translate to vector machine (SVM) algorithm and neural network.
99.78 % accuracy. Conventional method of fault detection and
classification are unreliable due to large power system
Keywords—Discrete wavelet transform, while wavelet transform in conjunction with
adaptive neuro-fuzzy inference system, fault type, AI/Fuzzy/expert system based techniques have the
fault detector, fault classifier, fault locator, feature advantage of fast response and increased accuracy
extraction as compared to conventional techniques [6, 7].
I. INTRODUCTION Previous work by other researchers presents
several methods. He, Fu, Lin and Boproposed wavelet
Transmission line is one of the most important singular entropy (WSE) technique which combines the
components in a power system. It constitutes the advantages of wavelet transform (WT), singular value
major part of power system. Transmission and decomposition (SVD) and Shannon entropy for fault
distribution lines are vital links between the generating detection and classification in transmission lines [8].
unit and consumers to achieve the continuity of He, Wu and Qian presented wavelet transform feature
electric supply. To economically transfer large blocks extraction method for automatic fault detection and
of power between systems and from remote classification in electrical power systems, and the
generating sites, high voltage (HV) and extra high wavelet singular entropy was shown to be sensitive to
voltage (EHV) overhead transmission systems are noise and abrupt variation of signal [9]. Gomes,
being used. Transmission lines also form a link in André, Costa, deFaria, and Caminhas used functional
interconnected system operation for bi-directional flow analysis and intelligent computing to detect and
of power. Transmission lines run over hundreds of classify the fault in transmission line of power system
kilometers to supply electrical power to the consumers [10]. Al-Kababjie, Al-Durzi and Al-Nuaimidescribed a
[1]. new distance relay for fault detection and
A fault occurs when two or more conductors come classification in transmission line that is used in
in contact with each other or ground in three phase wavelet transform [11]. Fault detection based on rate

www.jmest.org
JMESTN42353475 12367
 Journal of Multidisciplinary Engineering Science and Technology (JMEST)
ISSN: 2458-9403
Vol. 7 Issue 6, June - 2020

of change of frequency relay (ROCOF) and vector line. It aims at improving power efficiency by providing
surge (VS) relay was presented by Freitas, Xu, alternative and improved solutions to the challenges in
Affonso and Huang [12]. Chanda, Kishore and Sinha the conventional fault detection and classification
presented a method based on wavelet multi-resolution method for high tension transmission lines using
analysis (MRA) for fault classification [13]. discrete wavelet transform and adaptive neuro-fuzzy
Samantaray used S transform to calculate the inference system. The model will be developed and
statistical properties of the current signal and then the tested using a 132 kV transmission line.
tree decision was used as the input index of fuzzy
II. METHODOLOGY
logic to classify the fault [14]. IJayabharata and
Mohantadeveloped a wavelet multi-resolution analysis In this paper, empirical and simulation approaches
algorithm to classify fault in transmission lines [15]. is deployed in the development of the models as
The developed algorithm had a high accuracy summarized in Fig. 1. The data (current signals of the
irrespective of the fault location, fault resistance and three phases) obtained from the simulation of the
the fault inception angle. An intelligence technique or different fault conditions on the developed Simulink
automatic fault detection was presented in Srinivasan, model of 132 kV, Aba - Itu transmission lines (Fig. 2),
Cheu, Poh, and Chwee[16]. The method combined are first analyzed with DWT for features extraction.
fuzzy logic and genetic algorithms (GA). The extracted fault current features are then used to
train and develop an ANFIS models for the detection,
This paper presents yet another technique for the
classification of fault types and determining the actual
classification and detection of faults on a transmission
fault location.

Fig. 1: Overview of the fault detection, classification and location models development

Fig. 2: 132 kV, Aba – Itu transmission lines Simulink model for the fault data generation

www.jmest.org
JMESTN42353475 12368
 Journal of Multidisciplinary Engineering Science and Technology (JMEST)
ISSN: 2458-9403
Vol. 7 Issue 6, June - 2020

simulation. Fault current of the three phases were

The line parameters for the case study
used for the fault features extraction and analysis.
transmission lines are presented in Table 1, while the
parameter settings for generating fault values of B. Fault Features Extraction with DWT
currents for 132 kV, Aba – Itu transmission lines are in
The descriptions of the wavelet and associated
Table 2.
parameters employed are given in Table 3. The three
Table 1: Line parameters for 132 kV, Aba – Itu current signals for the different fault conditions were
transmission lines taken into consideration. In order to distinguish
Resistance Inductance Capacitance between L-L fault and L-L-G types of fault, zero
Line per unit per unit per unit sequence current was also taken into consideration.
Parameter length length length The root mean square (RMS) values of the details
component of the data window of each signal
(Ω/km) (mH/km) (nF/km)
represent the quantity of extracted features.
Positive
0.01273 0.9337 12.74 Table 3: Description of Wavelet and associated
Sequence
parameters
Zero
0.3864 4.1264 7.751 Parameters Properties
Sequence
Mother wavelet Doubechies, dB5
Table 2: Parameter settings for generation of fault Sampling frequency 20 kHz
data Information analyzed Details: d1 - d5
Parameter Set values Number of samples per cycle 400
L-G, L-L-G, L-L,L-L-L, Occurrence of fault Third cycle
Fault Type
No fault Data window length analyzed One cycle (20 ms)
Fault Location (km) 0.1 – 99.9
Fault Resistance (Ω) 0.1 – 2.5
C. ANFIS-based Fault Detection Model
Firing angle 0o – 180o
Fault Time (s) 0.1 The fault detection model was developed using
ANFIS as shown in Fig. 3. The inputs are the RMS
Load (MVA) 100
values of the entire detail components of phases A, B
Base Power (MVA) 100 and C of the transmission lines under consideration,
A. Generation of Fault Data while the output is a binary value which signifies the
state of the line (i.e. 0 – no fault, 1 – fault). Whenever
A total of 1211 simulations of different fault fault is confirmed to occur, a trip signal is sent to open
conditions at different locations were done at a the line circuit breakers. The 1211 sets of data
sampling rate of 20 kHz with a simulation time of 0.1 simulated are divided into three groups of 80 %
second. The fault was initiated at 0.05 second in each (training data), 10 % (testing data) and10 % (validation
data).

Fig. 3: Block diagram of ANFIS-based fault detection model

www.jmest.org
JMESTN42353475 12369
 Journal of Multidisciplinary Engineering Science and Technology (JMEST)
ISSN: 2458-9403
Vol. 7 Issue 6, June - 2020

D. ANFIS-based Fault Classification Model (CT) at a sampling frequency of 20 kHz. Whenever

The fault classification model was developed using the value of any of the phase currents exceeds the
ANFIS as shown in Fig. 4. The inputs are the RMS preset threshold (five times the rated current), discrete
values of the entire detail components of phases A, B, wavelet transform is applied on a data window frame
C and zero sequence current of the transmission lines of 400 samples by the feature extractor unit to obtain
under consideration, while the output is an integer the details components at level 1 to 5. The output of
which signifies the fault type. the extractor is the RMS values of the detail
components of the current signals which are fed into
E. ANFIS-based Fault Location Model the fault detector for the detection of the shunt fault. If
The fault location model was developed using any of the shunt faults is detected, a trip signal is sent
ANFIS as shown in Fig. 5. The first three inputs are the to trip the circuit breaker (CB) at busbar A.
RMS values of the entire detail components of phases
The fault classifier, on receiving notification of fault
A, B and C of the transmission lines under
from the detector, uses the RMS values of the current
consideration, while the last input is the type of fault.
of the three phases detail components as well as that
The output is the location of fault occurrence.
of the zero sequence current to determine the type of
F. DWT and ANFIS-Based Fault Classification and fault. While the fault locator uses the RMS values of
Location System the detail components of the current signals together
with the fault type to determine the fault location. Fig.
Fig. 6 shows the block diagram of the developed 7 present a flowchart that summarizes the working of
DWT and ANFIS-based system for fault detection,
the developed system.
classification and location on a 132 kV transmission
lines. The analogue-to-digital (ADC) discretized the
current signals obtained by the current transformer

Fig. 4: Block diagram of ANFIS-based fault classification model

Fig. 5: Block diagram of ANFIS-based fault location model

www.jmest.org
JMESTN42353475 12370
 Journal of Multidisciplinary Engineering Science and Technology (JMEST)
ISSN: 2458-9403
Vol. 7 Issue 6, June - 2020

Fig. 6: Single line diagram of the developed system

Fig. 7: DWT and ANFIS-based fault classification and location system flowchart

www.jmest.org
JMESTN42353475 12371
 Journal of Multidisciplinary Engineering Science and Technology (JMEST)
ISSN: 2458-9403
Vol. 7 Issue 6, June - 2020

III. RESULTS AND DISCUSSION developed fault detector was able to discriminate
between actual fault cases from normal condition. It is
The MATLAB/Simulink software was used for the
also observed that the maximum tripping time for all
implementation of DWT and ANFIS-based fault
fault occurrence is 8 milliseconds (Fig. 12). This mean
classification and location system. The current
that the fault is detected within one cycle of fault
waveform of a single phase to ground fault on phase A
occurrence.
in Fig. 8 shows that the value of the faulted phase
current exceeds the threshold value (5 p.u.). The ability for the DWT-ANFIS based classifier to
Comparing the DWT decomposition of the faulty phase accurately distinguish between the different fault types
(Fig. 9) to those of the other phases (Figs. 10 and 11) under different fault conditions is excellent as can be
shows that the faulty phase has far higher amplitude seen in Table 6.Some of simulation results from the
for all the detail levels (d1 – d5). This is further DWT-ANFIS based fault locator are also presented in
confirmed in Table 4. Table 6. It shows near accurate predictions of fault
distance in all cases. In Table 7, the R-squared values
Three phase fault was initiated at 0.05 second at a
of the predicted fault distances are approximately
distance of 10 km with generator firing angle of 90o
equal to 1. The maximum value of RMSE and absolute
and a fault resistance of 1 Ω. The output is either “0”
error for all fault types are 0.109 and 0.22 km
(no fault) or “1” (fault). As seen in Table 5, the
respectively.

Fig. 8: Current waveform of single phase to ground fault on line A

Fig. 9: DWT decomposition of phase A

www.jmest.org
JMESTN42353475 12372
 Journal of Multidisciplinary Engineering Science and Technology (JMEST)
ISSN: 2458-9403
Vol. 7 Issue 6, June - 2020

Fig. 10: DWT decomposition of phase B

Fig. 11: DWT decomposition of phase C

Table 4: RMS values of the detail components of the three phases for L-G fault
d1 d2 d3 d4 d5 Entire details (Ph_rms)

Phase A 0.0819 0.5228 1.3376 2.1774 4.4604 1.1587

Phase B 0.0019 0.0041 0.0103 0.0153 0.0354 0.0090
Phase C 0.0019 0.0040 0.0090 0.0125 0.0286 0.0075

www.jmest.org
JMESTN42353475 12373
 Journal of Multidisciplinary Engineering Science and Technology (JMEST)
ISSN: 2458-9403
Vol. 7 Issue 6, June - 2020

Table 5: Fault detection output for different fault scenarios

Fault RMS Values Fault Detector Tripping
Fault Type
Time (sec) Phase A Phase B Phase C (output) Time (sec)
Normal - 0.0024 0.0082 0.0067 0 -
A-G 0.05 1.1587 0.0090 0.0075 1 0.0576
B-G 0.05 0.0032 0.6875 0.0070 1 0.0576
C-G 0.05 0.0029 0.0085 0.5696 1 0.0576
AB 0.05 1.3729 1.3732 0.0067 1 0.0576
BC 0.05 0.0024 0.3808 0.3800 1 0.058
AC 0.05 1.2698 0.0082 1.2701 1 0.0576
AB-G 0.05 1.5527 1.2047 0.0069 1 0.0576
BC-G 0.05 0.0037 0.6400 0.5061 1 0.0576
AC-G 0.05 1.5035 0.085 1.0472 1 0.0576
ABC 0.05 1.7450 1.0429 0.8511 1 0.0576

Fig. 12: Current waveform of three phase shunt fault with the trip signal sent to the circuit breaker

www.jmest.org
JMESTN42353475 12374
 Journal of Multidisciplinary Engineering Science and Technology (JMEST)
ISSN: 2458-9403
Vol. 7 Issue 6, June - 2020

Table 6: Fault classification and location output for different fault scenarios
Fault Type @ Actual Predicted Fault Type
S/N |error| (%)
Fault Resistance Distance Distance Classifier
1 AG @ 2.25 Ω 0.1 0.133 0.033 AG √
2 AG @ 1 Ω 50.0 50.008 0.008 AG √
3 AG @ 2.5 Ω 90.0 89.967 0.033 AG √
4 BG @ 1.5 Ω 20.0 19.988 0.012 BG √
5 BG @ 0.1 Ω 80.0 80.004 0.004 BG √
6 BG @ 2 Ω 99.9 99.874 0.026 BG √
7 CG @ 1 Ω 30.0 30.045 0.045 CG √
8 CG @ 0.25 Ω 60.0 59.987 0.013 CG √
9 CG @ 2 Ω 90.0 89.944 0.056 CG √
10 AB @ 1.25 Ω 10.0 9.993 0.007 AB √
11 AB @ 0.5 Ω 40.0 40.000 0.000 AB √
12 AB @ 2.25 Ω 70.0 70.012 0.012 AB √
13 AC @ 0.75 Ω 20.0 20.030 0.030 AC √
14 AC @ 0.1 Ω 50.0 49.970 0.030 AC √
15 AC @ 2 Ω 70.0 70.002 0.002 AC √
16 BC @ 1 Ω 0.1 0.171 0.071 BC √
17 BC @ 0.75 Ω 10.0 9.948 0.052 BC √
18 BC @ 0.1 Ω 40.0 40.027 0.027 BC √
19 ABG @ 0.25 Ω 20.0 19.866 0.134 ABG √
20 ABG @ 1.75 Ω 60.0 59.976 0.024 ABG √
21 ABG @ 0.75 Ω 99.9 99.836 0.064 ABG √
22 ACG @ 2.25 Ω 30.0 30.134 0.134 ACG √
23 ACG @ 1.25 Ω 70.0 69.929 0.071 ACG √
24 ACG @ 1 Ω 80.0 80.018 0.018 ACG √
25 BCG @ 2.5 Ω 10.0 9.940 0.060 BCG √
26 BCG @ 1.5 Ω 50.0 50.000 0.000 BCG √
27 BCG @ 0.75 Ω 80.0 80.078 0.078 BCG √
28 ABC @ 2.5 Ω 0.1 0.126 0.026 ABC √
29 ABC @ 1.5 Ω 40.0 40.047 0.047 ABC √
30 ABC @ 0.5 Ω 80.0 79.953 0.047 ABC √

Table 7:Performance evaluation of DWT-ANFIS based system for fault location

S/N R-squared RMSE |error|
1 1.00000 0.0232 0 - 0.037
2 1.00000 0.0418 0.002 - 0.087
3 1.00000 0.0378 0 - 0.07
4 1.00000 0.0228 0 - 0.031
5 1.00000 0.0252 0.002 - 0.035
6 1.00000 0.0451 0 - 0.071
7 0.99999 0.1091 0.009 - 0.221
8 0.99999 0.0796 0.007 - 0.163
9 1.00000 0.0411 0 - 0.078
10 1.00000 0.0455 0.01 - 0.077

www.jmest.org
JMESTN42353475 12375
 Journal of Multidisciplinary Engineering Science and Technology (JMEST)
ISSN: 2458-9403
Vol. 7 Issue 6, June - 2020

IV. CONCLUSION International Journal of Electrical and Computer

Engineering, 2(10), 2389 - 2393.
Transmission lines are highways for the transfer of
electric energy from generating stations through a long [7] Kim, C. H., Kim, H., Ko, Y. H., Byun, S. H.,
distance to distribution stations. These lines are Aggrawal, R. K., and Johns, A. T. (2002): A novel
sometimes exposed to severe atmospheric conditions fault detection technique of high impedance
and unfavourable terrains, thus the likelihood of fault arcing faults in transmission lines using the
occurrence is very high. Under fault condition, power wavelet transform. In IEEE Transactions on
system equipment is subjected to the greatest stress Power Delivery, 17(4): 921 - 929
which can damage the equipment as well as affect the [8] He, Z., Fu, L., Lin, S., and Bo, z. (2010): Fault
quality and stability of the power. It is necessary to detection and classification in EHV transmission
ensure that the fault is cleared as quickly as possible. line based on wavelet singular entropy. Power
In this work, a reliable scheme for the detection, Delivery, IEEE Transactions on 25(4): 2156-
classification and location of fault on transmission lines 2163.
is developed. This scheme combines the feature
[9] He, Z., Wu, X., and Qian, Q. (2004): Automatic
extraction capability of the discrete waveform
fault detection for power system using wavelet
transform and the intelligent classification capability of
singular entropy. International Conference on
adaptive neuro-fuzzy inference system. The developed
Intelligent Mechatronics and Automation
DWT-ANFIS model was tested and the results
Proceedings, 433-437.
compared with Impedance-ANFIS model. Faults were
detected within 8ms that is less than one complete [10] Gomes, D. S., André, Costa, M. A., deFaria, T.
cycle from fault inception to prevent equipment G. A., and Caminhas, W. M. (2013): Detection
damage and prolonged power outage. An accuracy and classification of faults in power transmission
level of 99.78% and a RMSE of 0.0228-0.1091 were lines using functional analysis and computational
obtained. intelligence.In IEEE Transactions on Power
Delivery, 28(3): 1402 - 1413
REFERENCES
[11] Al-Kababjie, M. F., Al-Durzi, F., and Al-Nuaimi, N.
[1] Patel, M. and Patel, R. N. (2012): Fault detection H. (2012): A fault detection and classification
and classification on a transmission line using using new distance relay. 2012 First International
wavelet multi resolution analysis and neural Conference on Renewable Energies and
network.International Journal of Computer Vehicular Technology, 237-243
Applications, 47(22): 27-33.
[12] Freitas, W., Xu, W., Affonso, C. M., and Huang,
[2] Taywade, D. D. and Ghute, A. A. (2016): Z. (2005). Comparative analysis between
Detection and classification of transmission lines ROCOF and vector surge relays for distributed
faults using discrete wavelet transform and ANN generation applications. In IEEE Transactions on
as classifier. International Journal of Engineering Power Delivery,20(2): 1315-1324.
Sciences & Research Technology, 5(12): 924-
[13] Chanda, D., Kishore, N. K., and Sinha, A. K.
933.
(2003). Application of wavelet multi-resolution
[3] Hatata, A. Y., Hassan, Z. M. and Eskander, S. S. analysis for classification of faults on
(2016): Transmission Line protection scheme for transmission lines,TENCON 2003. Conference
fault detection, classification and location using on Convergent Technologies for Asia-Pacific, 4:
ANN.InternationalJournal of Modern Engineering 1464-1469.
Research, 6(8): 1-10.
[14] Samantaray, S. R. (2013). Systematic fuzzy rule
[4] Nayeripour, M., Rajaei, A. H., Ghanbarian, M. M based approach for fault classification in
and Dehghani, M. (2015): Fault detection and transmission lines. Applied Soft Computing,13(2):
classification in transmission lines based on a 928-938.
combination of wavelet singular values and fuzzy
[15] Jayabharata, R. M., and Mohanta, D. K. (2007).
logic. The Second National Conference on
A wavelet-fuzzy combined approach for
Applied Research in Science and Technology,
classification and location of transmission line
36(3): 69-82.
faults. International Journal of ElectricalPower &
[5] Thokala, K. R. and Naik, B. B. (2017): Fault Energy Systems, 29(9): 669-678.
detection and classification on a high voltage
[16] Srinivasan, D., Cheu, R. L., Poh, Y. P., and
transmission line using wavelet transforms.
Chwee, A. K. (2000). Automated fault detection in
International Journal for Research in Applied
power distribution networks using a hybrid fuzzy–
Science & Engineering Technology, 5(9): 399-
genetic algorithm approach. Engineering
411.
Applications of Artificial Intelligence, 13(4): 407-
[6] Kale, V. S., Bhide, S. R., Bedekar, P. P., and 418.
Mohan, G. V. K. (2008): Detection and
classification of faults on parallel transmission
lines using wavelet transform and neural network.

www.jmest.org
JMESTN42353475 12376