Engineering, Technology & Applied Science Research Vol. 12, No.

4, 2022, 8972-8977 8972

An Intelligent Fault Detection and Classification

Scheme for Distribution Lines Using Machine
Learning

Balamurali Krishna Ponukumati Pampa Sinha

School of Electrical Engineering School of Electrical Engineering
KIIT Deemed to be University KIIT Deemed to be University
Bhubaneswar, India and Bhubaneswar, India
Department of EEE pampa.sinhafel@kiit.ac.in
Madanapalle Institute of Technology and Science
Madanapalle, India
pbmk20031993@gmail.com

Manoj Kumar Maharana A. V. Pavan Kumar

School of Electrical Engineering Department of EEE
KIIT Deemed to be University Madanapalle Institute of Technology and Science
Bhubaneswar, India Madanapalle, India
mkmfel@kiit.ac.in pawanrao82@gmail.com

Akkenaguntla Karthik
Department of EEE
Madanapalle Institute of Technology and Science
Madanapalle, India
akarthik160@gmail.com

Received: 28 May 2022 | Revised: 8 June 2022 | Accepted: 9 June 2022

Abstract-The current paper focuses on the development and I. INTRODUCTION

deployment of Machine Learning (ML) based algorithms for the
classification and detection of different faults in the electrical The intricacy of power system networks presents many
distribution system. The methodology adapted using ML has difficulties in distributing and preserving electricity. The
higher computational accuracy than traditional computational primary aim of the protection system is safeguarding the
algorithms. The parameters involved in developing ML for fault overall system by identifying faults and preventing
detection/classification are fundamental frequency, fault voltage, malfunctions to attain adequate power supply continuity [1].
and current components at fault situations. During faults, the Numerous domestic and commercial clients possess equipment
current and voltage waveforms consist of high-frequency that is susceptible to electrical interference [2]. A good
transient signals. The Wavelet Decomposition (WD) technique is protection system must ensure system stability by protecting
used to break down transient signals to obtain the required against all types of failures at all network locations. A system
information. To investigate the performance of the ML-based built with the current technology must be trustworthy and
algorithms, an IEEE 33 bus system is utilized, and a fault is dependable [3]. Phasor Measurement Units (PMUs) can detect
generated in Matlab/Simulink environment. The methodologies a variety of typical distribution system events, including load
used for fault detection and classification are K Nearest Neighbor and equipment switching transients and short circuit faults [4,
(KNN), Decision Tree (DT), and Support Vector Machine (SVM).
5].
The performance of the designed algorithm is assessed by
employing a confusion matrix, and the results demonstrated Convolutional Neural Networks (CNNs) are implemented
extraordinarily high accuracy. in a variety of fields, including object recognition and language
recognition. CNNs are getting increasingly popular, notably in
Keywords-machine learning; wavelet decomposition; k nearest power system management and control, due to their capacity
neighbor; decision tree; support vector machine and feasibility in identifying and classifying issues that arise in
Corresponding author: Balamurali Krishna Ponukumati
www.etasr.com Ponukumati et al.: An Intelligent Fault Detection and Classification Scheme for Distribution Lines …
 Engineering, Technology & Applied Science Research Vol. 12, No. 4, 2022, 8972-8977 8973

power systems [6-11]. The authors of [12] employed Support merely a few characteristics to be adjusted, and have many
Vector Regression (SVR) to enhance the effectiveness of short- applications in real-world situations, but the training procedure
term estimation techniques based on historical electricity is rather hard for high-dimension challenges. The gradient-
demand information. Authors in [13] suggest fault based back propagation approach for non-linear discrete
categorization in a distribution network by implementing an classification tasks provides a locally optimal solution. Slow
Adaptive Neuro-Fuzzy Inference System (ANFIS), and authors convergence is provided by ANNs in the back propagation
in [14] discuss rapid fault recognition and categorization in method. The convergence is determined by the initial value of
electrical distribution networks employing a hybrid of fuzzy weight restrictions linked to the system. Conventional relay
logic and singular entropy theory. The application of ML is not systems use mechanical parts to detect a fault. These
restricted to limited fields. It is also applied to the agriculture mechanical parts need periodical maintenance, otherwise their
sector in detecting plant diseases [15]. The necessity for a performance will be degraded and it should match the ratings
highly precise and resilient waveform deviation tracking of the transmission or distribution line.
system in high-voltage power systems makes research into its
categorization and identification using deep learning techniques Figure 1 depicts the flowchart of the suggested approach,
a popular subject [11, 16, 17]. In [18], a hybrid strategy of where Va, Vb, and Vc are the three-phase voltages, Ia, Ib, and Ic
Gated Recurrent Unit (GRU) with Discrete Wavelet Transform are the three-phase currents, VCOFF and ICOFF are Voltage and
(DWT) involved features employed as inputs. Fault recognition current coefficients, and the angles between V and I are
and categorization in microgrid employing CNNs was considered as inputs. These inputs are fed to the developed ML
proposed in [19] detection and classification algorithm which continuously
monitors the status of inputs until a fault occurs.
Fast Fourier Transform (FFT), DWT, S-Transform (ST)
and statistical techniques are some examples [20-23] of ML
algorithms used for fault detection and classification in
distributions lines and microgrids. Usually, the extraction of the
feature components is specific to the system setup and might
involve continued attempts, limiting generalization [24]. DWT
is widely used in the feature extraction step [25]. However, the
feature extraction section and functioning approaches are
unique throughout respective deployments. For example, as a
functional technique for defect identification, [26] employs
DWT calculated singular entropy including a set of parameters. Fig. 1. Flow chart of the proposed methodology.
This approach works perfectly, however, the margin to greater
fault impedance wasn’t addressed. In [27], DWT is employed II. FAULT DETECTION ANALYSIS
to collect features from current signals, while Particle Swarm
Optimization (PSO) is used to select the best wavelet WD approach is considered to detect the fault and classify
configuration. For fault categorization and diagnosis in a it. The WD method was also utilized in [29, 30] to extract the
microgrid, authors in [10, 25] propose ML and in [18] propose waveform but that’s a 3-level decomposition, whereas this
deep learning-based methods. To categorize the problem, the approach uses a 6-level WD as shown in Figure 2. The
wavelet transform is utilized to extract data from current fault/disturbance signal is routed via Low Pass Filters (LPFs)
signals, which are then classified using SVM [28] and DT [12]. and High Pass Filters (HPFs) until the necessary signal for
analysis is achieved. The fault/disturbance signal is
Techniques such as ANFIS, Fuzzy Interference Systems decomposed at 6 levels by passing via an LPF D and an HPF
(FIS), Artificial Neural Networks (ANNs), Feedforward Neural A. At level 1, the signal is decimated by 2 and D1 and A1
Networks (FNNs), CNNs, and conventional fault identification signals are obtained. D2 and A2 are obtained by passing the A1
with relay systems are some efficient ML techniques. ML signal through the second stage of LPF and HPF. Again, A2
focuses on extracting knowledge from a large set of passes through filters until level 6 and the final obtained signals
information and provides conventional ways for problem- are D6 and A6. As a result, significant information is derived
solving. It really can deal with a wide range of information and from the initial signal which has to be evaluated.
be employed in various situations. Furthermore, by integrating
more data for training, the accuracy of ML algorithms III. PHASE LOCKED LOOP
improves. Controllers like Fuzzy logic and FIS are A Phase-Locked Loop (PLL) is a regulatory technique
straightforward and use "if-then" relationships to handle which produces a signal as output with a phase-matched to that
problems with uncertainty. They are a generally strong of an input signal. Its block diagram is illustrated in Figure 3.
candidate even though they require huge training data sets. In order to keep the inlet and outlet phases locked, the input
Incorporating ANFIS assists with tunable characteristics, and output frequencies must be steady. Using the input voltage,
decreases computational time, gives quick convergence and Vabc, Vd, and Vq are obtained. Vd and Vq are voltage values near
responsiveness, but the difficulty arises when large the Point of Common Coupling (PCC) in the d- and q- axes
computation is expected, which is removed by adopting ML. respectively. The output is an approximation of the voltage
The ANNs, FNNs, and CNNs help predict the specific fault phase. The suggested PLL architecture for fault identification is
type and are simple to apply. Their use is simple, involving depicted in Figure 3.

www.etasr.com Ponukumati et al.: An Intelligent Fault Detection and Classification Scheme for Distribution Lines …
 Engineering, Technology & Applied Science Research Vol. 12, No. 4, 2022, 8972-8977 8974

Fig. 2. Six level wavelet-decomposition.

IV. MACHINE LEARNING

As the name implies, ML is a phrase that is used to define
the field of learning algorithms. An ML model will generally
have three different phases: training, validation, and real-time
implementation. During the training process, the system is fed
through test data with known correlations so that it can detect
patterns and create predictions within the same set accurately.
The validation stage is intended to assess the overall system
Fig. 3. Block diagram of the PLL. performance utilizing similar but disjoint data, as well as
previously identified responses. In terms of model training
In comparison to the block diagram in Figure 2, the error methods, ML is typically classified into four major
signal e(t) relates to the waveform deviation, while the methodologies: supervised, unsupervised, semi-supervised, and
Proportional-Integral (PI) regulator and integrator correlate to reinforced learning. Models which are trained on a sample with
the loop filter and Voltage Controlled Oscillator (VCO) knowing labelled outcomes are referred to as supervised
respectively. The three-phase voltages (Va, Vb, and Vc) that are learning. A list of classification algorithms that have been
sent into the PLL are first changed to αβ quantities (Vα and Vβ) frequently used in the literature follows.
using Clarke’s transformation. A schematic of a PQ-PLL block
is demonstrated in Figure 4. The quantities are then transferred A. Support Vector Machine
into a reference frame. Considering the proximity of the αβ
During appropriate post-fault diagnosis, a knowledge-based
quantities to the reference frame, an error signal is created. The
approach that is based on SVMs is offered. SVMs are used as
error signal is sent into a PI regulator, which controls the error
an intelligence tool to detect the faulty line originating first
to zero. The input signals are in phase with the frame of
from the substation and to determine the distance from that.
reference once the error is zero. When the PLL's output angle is
SVMs are compared to radial-based neural networks in datasets
in phase with the current of phase A, the error is exactly zero.
that depict diverse transmission system challenges.
When there is a transient in the system, the error signal deviates
from zero. The magnitude and frequency of the deviation will B. Decision Tree
vary depending on the features of the transient.
The DT's design is simple, and we can easily follow the tree
structure to describe how to make a conclusion. The vast scope
of power system DTs has lately been discovered to be
extremely effective in applications such as online dynamic
safety evaluation, stability to transients, and islanding
identification.
C. K Nearest Neighbor
The KNN approach is a safe, supervised ML approach
utilized to tackle classification and regression issues. Faults
may be detected and recognized in distance protection using
Fig. 4. Block diagram of the PQ-PLL. the KNN method.

www.etasr.com Ponukumati et al.: An Intelligent Fault Detection and Classification Scheme for Distribution Lines …
 Engineering, Technology & Applied Science Research Vol. 12, No. 4, 2022, 8972-8977 8975

D. Bayesian Learner B. Case Study

The Bayesian classifiers, also called Naïve Bayes are The developed ML algorithms are tested on the IEEE 33
statistical classifiers that estimate the probability of class bus system (Figure 5). The suggested fault detection and
membership using supervised learning techniques. classification technique was tested using the IEEE 33 radial
distribution scheme.
E. Sequential Minimal Optimization (SMO)
A fault is developed to analyze the performance of the
Training an SVM necessitates the solving of a large system. Figure 6 illustrates the occurrence of an L-G fault
quadratic programming optimization problem. SMO divides between 0.2s and 0.4s in a three-phase voltage waveform. WD
this massive quadratic programming issue into a bunch of is used to assess the fault voltage signal. Figure 7 depicts the L-
smaller quadratic programming problems. G fault voltage signal of a multiresolution WD. The Db6
F. Logistic Regression wavelet is used for producing a 6-step decomposition. At every
frequency spectrum, the rebuilt variants of every element,
If the dependent variable is binary, logistic regression is the along with the approximation and records of the unique signal,
best regression approach to use. Like other regression studies, are represented.
it is a statistical approach. The dependent variable Y in logistic
regression has values of 1 and 0 for the outcomes of interest.
G. Linear Regression
The relation between both the input variables (x) and output
variables (y) for linear regression is generally stated as a linear
equation of the form y = a + bx. As a result, the aim of linear
regression would be to determine the values of coefficients a
and b, where a is the interception and b is the line's slope.
V. RESULTS AND DISCUSSION
A. The Considered ML Techniques Fig. 5. The IEEE 33 bus system.
To detect the fault and classify it in the distribution
network, the testing data are evaluated using SVM and DT
algorithms.
• If there are little training data (less than 1000), support
vector methods are often good because they are highly
regularized.
• SVMs have better production results than ANNs, while
they avoid the problem of over-fitting. One can use the
kernel function in the SVM to solve complex problems.
They can effectively handle data with higher dimensions
that are linear inseparable.
For data preprocessing, WD is employed. Matlab is used to
obtain various operating and fault data. A total of 2746 input Fig. 6. Three-phase voltage waveforms under L-G fault occurrence.
samples are considered, with 70% are used for training and
30% for testing the model’s performance. The training data are C. Confusion Matrix
used to train a model that can predict how to respond to
incoming new data. Table I shows the accuracy rates of the The confusion matrix is the most often used method for
tested algorithms. The cubic SVM algorithm yielded the best evaluating the accuracy and correctness of a classification
accuracy rate of 87.1%. algorithm. It also helps identifying areas where the classifier
has underperformed. It can also help understanding what the
TABLE I. PERFORMANCE ANALYSIS classification model is doing correctly and what sorts of errors
it is producing. In Figure 8, the faulty classification confusion
Classifier Accuracy
Fine tree 82.20%
matrix is shown. It is evident that the suggested system can
Medium tree 79.40% classify various faults, and the prediction rate is high. Figure 9
Coarse tree 55.20% shows the True Positive Rate (TPR) and False Negative Rate
Linear SVM 52.50% (FNR) data and the TPR value is high, which says that the
Quadratic SVM 78.20% model can detect faults accurately.
Cubic SVM 87.10%
Fine Gaussian SVM 63.60% Precision is a measure of how accurate a classifier is. A low
Medium Gaussian SVM 67.20% accuracy might also be indicative of a massive amount of False
Coarse Gaussian SVM 51.80% Positives. Figure 10 shows the Positive Predictive Value (PPV)

www.etasr.com Ponukumati et al.: An Intelligent Fault Detection and Classification Scheme for Distribution Lines …
 Engineering, Technology & Applied Science Research Vol. 12, No. 4, 2022, 8972-8977 8976

and False Discovery Rate (FDR) values. It is clear that the PPV
has higher value than FDR. The rows of the confusion matrix
correspond to the anticipated class, while the columns are
commensurate to the actual class. The diagonal cells are
commensurate to suitably categorized observations. The off-
diagonal cells are commensurate to observations that were
wrongly categorized.

Fig. 10. PPV and FDR values of data.

VI. CONCLUSION
The utilization of ML algorithms for fault categorization
and identification on a three-distribution network was
extensively investigated in this study. The created systems used
instantaneous current and voltage data as inputs for fault
detection and categorization. This paper described a waveform-
based fault classification method for electrical distribution
systems. WD was employed for bringing out the features from
recorded wave data. The massive volume of wave data
processing activity was treated effectively by carefully
spreading the computing burden. ML modules identified and
classified the faults. The fault classification and detection
results demonstrate the efficacy of ML in categorizing and
identifying faults in less time than other traditional relaying
Fig. 7. WD analysis on LG fault.
systems. The systems make use of the natural capacity for
pattern classification and identification. For distance protection,
ML technology is strong, effective, reliable and precise.
REFERENCES
[1] H. Zayandehroodi, A. Mohamed, H. Shareef, and M. Mohammadjafari,
"An automated protection method for distribution networks with
distributed generations using radial basis function neural network," in
5th International Power Engineering and Optimization Conference,
Shah Alam, Malaysia, Jun. 2011, pp. 255–260, https://doi.org/10.1109/
PEOCO.2011.5970384.
[2] F. Jandan, S. Khokhar, Z. A. Memon, and S. A. A. Shah, "Wavelet
Based Simulation and Analysis of Single and Multiple Power Quality
Disturbances," Engineering, Technology & Applied Science Research,
vol. 9, no. 2, pp. 3909–3914, Apr. 2019, https://doi.org/10.48084/etasr.
2409.
[3] P. Jafarian and M. Sanaye-Pasand, "A Traveling-Wave-Based Protection
Fig. 8. Faulty classification confusion matrix. Technique Using Wavelet/PCA Analysis," IEEE Transactions on Power
Delivery, vol. 25, no. 2, pp. 588–599, Apr. 2010, https://doi.org/10.1109/
TPWRD.2009.2037819.
[4] R. Yadav, S. Raj, and A. K. Pradhan, "Real-Time Event Classification in
Power System With Renewables Using Kernel Density Estimation and
Deep Neural Network," IEEE Transactions on Smart Grid, vol. 10, no.
6, pp. 6849–6859, Aug. 2019, https://doi.org/10.1109/TSG.2019.
2912350.
[5] A. Shahsavari, M. Farajollahi, E. M. Stewart, E. Cortez, and H.
Mohsenian-Rad, "Situational Awareness in Distribution Grid Using
Micro-PMU Data: A Machine Learning Approach," IEEE Transactions
on Smart Grid, vol. 10, no. 6, pp. 6167–6177, Aug. 2019, https://doi.org/
10.1109/TSG.2019.2898676.
[6] A. Bagheri, I. Y. H. Gu, M. H. J. Bollen, and E. Balouji, "A Robust
Transform-Domain Deep Convolutional Network for Voltage Dip
Classification," IEEE Transactions on Power Delivery, vol. 33, no. 6,
pp. 2794–2802, Sep. 2018, https://doi.org/10.1109/TPWRD.2018.
Fig. 9. TPR and FNR values of data.
2854677.

www.etasr.com Ponukumati et al.: An Intelligent Fault Detection and Classification Scheme for Distribution Lines …
 Engineering, Technology & Applied Science Research Vol. 12, No. 4, 2022, 8972-8977 8977
[7] S. Lan, M.-J. Chen, and D.-Y. Chen, "A Novel HVDC Double-Terminal [22] M. H. H. Musa, Z. He, L. Fu, and Y. Deng, "A correlation coefficient-
Non-Synchronous Fault Location Method Based on Convolutional based algorithm for fault detection and classification in a power
Neural Network," IEEE Transactions on Power Delivery, vol. 34, no. 3, transmission line," IEEJ Transactions on Electrical and Electronic
pp. 848–857, Jun. 2019, https://doi.org/10.1109/TPWRD.2019.2901594. Engineering, vol. 13, no. 10, pp. 1394–1403, 2018, https://doi.org/
[8] Y. Liu, S. Pei, W. Fu, K. Zhang, X. Ji, and Z. Yin, "The discrimination 10.1002/tee.22705.
method as applied to a deteriorated porcelain insulator used in [23] B. K. Ponukumati, P. Sinha, M. K. Maharana, C. Jenab, A. V. Pavan
transmission lines on the basis of a convolution neural network," IEEE Kumar, and K. Akkenaguntla, "Pattern Recognition Technique Based
Transactions on Dielectrics and Electrical Insulation, vol. 24, no. 6, pp. Fault Detection in Multi-Microgrid," in 2nd International Conference on
3559–3566, Sep. 2017, https://doi.org/10.1109/TDEI.2017.006840. Applied Electromagnetics, Signal Processing, & Communication,
[9] S. R. Fahim, Y. Sarker, S. K. Sarker, Md. R. I. Sheikh, and S. K. Das, Bhubaneswar, India, Nov. 2021, pp. 1–6, https://doi.org/10.1109/
"Self attention convolutional neural network with time series imaging AESPC52704.2021.9708541.
based feature extraction for transmission line fault detection and [24] K. Chen, J. Hu, and J. He, "Detection and Classification of Transmission
classification," Electric Power Systems Research, vol. 187, Oct. 2020, Line Faults Based on Unsupervised Feature Learning and Convolutional
Art. no. 106437, https://doi.org/10.1016/j.epsr.2020.106437. Sparse Autoencoder," IEEE Transactions on Smart Grid, vol. 9, no. 3,
[10] J. Han, S. Miao, Y. Li, W. Yang, and H. Yin, "Faulted-Phase pp. 1748–1758, Feb. 2018, https://doi.org/10.1109/TSG.2016.2598881.
classification for transmission lines using gradient similarity [25] D. P. Mishra, S. R. Samantaray, and G. Joos, "A Combined Wavelet and
visualization and cross-domain adaption-based convolutional neural Data-Mining Based Intelligent Protection Scheme for Microgrid," IEEE
network," Electric Power Systems Research, vol. 191, Feb. 2021, Art. Transactions on Smart Grid, vol. 7, no. 5, pp. 2295–2304, Sep. 2016,
no. 106876, https://doi.org/10.1016/j.epsr.2020.106876. https://doi.org/10.1109/TSG.2015.2487501.
[11] S. Barrios, D. Buldain, M. P. Comech, I. Gilbert, and I. Orue, "Partial [26] Z. He, L. Fu, S. Lin, and Z. Bo, "Fault Detection and Classification in
Discharge Classification Using Deep Learning Methods—Survey of EHV Transmission Line Based on Wavelet Singular Entropy," IEEE
Recent Progress," Energies, vol. 12, no. 13, Jan. 2019, Art. no. 2485, Transactions on Power Delivery, vol. 25, no. 4, pp. 2156–2163, Jul.
https://doi.org/10.3390/en12132485. 2010, https://doi.org/10.1109/TPWRD.2010.2042624.
[12] N. T. Dung and N. T. Phuong, "Short-Term Electric Load Forecasting [27] T. S. Abdelgayed, W. G. Morsi, and T. S. Sidhu, "A New Approach for
Using Standardized Load Profile (SLP) And Support Vector Regression Fault Classification in Microgrids Using Optimal Wavelet Functions
(SVR)," Engineering, Technology & Applied Science Research, vol. 9, Matching Pursuit," IEEE Transactions on Smart Grid, vol. 9, no. 5, pp.
no. 4, pp. 4548–4553, Aug. 2019, https://doi.org/10.48084/etasr.2929. 4838–4846, Sep. 2018, https://doi.org/10.1109/TSG.2017.2672881.
[13] J. Zhang, Z. Y. He, S. Lin, Y. B. Zhang, and Q. Q. Qian, "An ANFIS- [28] M. Manohar and E. Koley, "SVM based protection scheme for
based fault classification approach in power distribution system," microgrid," in International Conference on Intelligent Computing,
International Journal of Electrical Power & Energy Systems, vol. 49, Instrumentation and Control Technologies, Kerala, India, Jul. 2017, pp.
pp. 243–252, Jul. 2013, https://doi.org/10.1016/j.ijepes.2012.12.005. 429–432, https://doi.org/10.1109/ICICICT1.2017.8342601.
[14] M. Dehghani, M. H. Khooban, and T. Niknam, "Fast fault detection and [29] P. Balamurali Krishna and P. Sinha, "Detection of Power System
classification based on a combination of wavelet singular entropy theory Harmonics Using NBPSO Based Optimally Placed Harmonic
and fuzzy logic in distribution lines in the presence of distributed Measurement Analyser Units," in Second International Conference on
generations," International Journal of Electrical Power & Energy Computing Methodologies and Communication, Erode, India, Feb. 2018,
Systems, vol. 78, pp. 455–462, Jun. 2016, https://doi.org/10.1016/ pp. 369–373, https://doi.org/10.1109/ICCMC.2018.8488114.
j.ijepes.2015.11.048. [30] B. K. P, P. Sinha, M. K. Maharana, C. Jena, A. V. Pavan Kumar, and K.
[15] S. Nuanmeesri, "A Hybrid Deep Learning and Optimized Machine Akkenaguntla, "Power System Fault Detection Using Image Processing
Learning Approach for Rose Leaf Disease Classification," Engineering, And Pattern Recognition," in 2nd International Conference on Applied
Technology & Applied Science Research, vol. 11, no. 5, pp. 7678–7683, Electromagnetics, Signal Processing, & Communication, Bhubaneswar,
Oct. 2021, https://doi.org/10.48084/etasr.4455. India, Nov. 2021, pp. 1–5, https://doi.org/10.1109/AESPC52704.2021.
[16] S. Lu, H. Chai, A. Sahoo, and B. T. Phung, "Condition Monitoring 9708475.
Based on Partial Discharge Diagnostics Using Machine Learning
Methods: A Comprehensive State-of-the-Art Review," IEEE
Transactions on Dielectrics and Electrical Insulation, vol. 27, no. 6, pp.
1861–1888, Dec. 2020, https://doi.org/10.1109/TDEI.2020.009070.
[17] G. Li, X. Wang, X. Li, A. Yang, and M. Rong, "Partial Discharge
Recognition with a Multi-Resolution Convolutional Neural Network,"
Sensors, vol. 18, no. 10, Oct. 2018, Art. no. 3512, https://doi.org/
10.3390/s18103512.
[18] J. J. Q. Yu, Y. Hou, A. Y. S. Lam, and V. O. K. Li, "Intelligent Fault
Detection Scheme for Microgrids With Wavelet-Based Deep Neural
Networks," IEEE Transactions on Smart Grid, vol. 10, no. 2, pp. 1694–
1703, Mar. 2019, https://doi.org/10.1109/TSG.2017.2776310.
[19] M. Manohar, E. Koley, S. Ghosh, D. K. Mohanta, and R. C. Bansal,
"Spatio-temporal information based protection scheme for PV integrated
microgrid under solar irradiance intermittency using deep convolutional
neural network," International Journal of Electrical Power & Energy
Systems, vol. 116, Mar. 2020, Art. no. 105576, https://doi.org/
10.1016/j.ijepes.2019.105576.
[20] K. Chen, C. Huang, and J. He, "Fault detection, classification and
location for transmission lines and distribution systems: a review on the
methods," High Voltage, vol. 1, no. 1, pp. 25–33, 2016, https://doi.org/
10.1049/hve.2016.0005.
[21] P. Bunnoon, "Fault Detection Approaches to Power System: State-of-
the-Art Article Reviews for Searching a New Approach in the Future,"
International Journal of Electrical and Computer Engineering, vol. 3,
no. 4, pp. 553–560, May 2013, https://doi.org/10.11591/ijece.v3i4.3195.

www.etasr.com Ponukumati et al.: An Intelligent Fault Detection and Classification Scheme for Distribution Lines …