International Journal of Power Electronics and Drive Systems (IJPEDS)

Vol. 15, No. 3, September 2024, pp. 1851~1870

ISSN: 2088-8694, DOI: 10.11591/ijpeds.v15.i3.pp1851-1870  1851

CNN based fault event classification and power quality

enhancement in hybrid power system

Abdul Quawi1, Y. Mohamed Shuaib1, M. Manikandan2

1
Department of Electrical and Electronics Engineering, B. S. Abdur Rahman Crescent Institute of Science and Technology,
Chennai, India
2
Department of Electrical and Electronics Engineering, Jyothishmathi Institute of Technology and Science, Telangana, India

Article Info ABSTRACT

Article history: A resilient approach is presented in this study for detecting and classifying
faults for power distribution systems integrating renewable energy sources
Received Aug 26, 2023 (RES). Combining discrete wavelet transform (DWT) and convolutional
Revised Feb 29, 2024 neural network (CNN). The suggested framework addresses the challenges of
Accepted Mar 28, 2024 RES intermittency and kinetic energy insufficiency. The recommended
methodology is evaluated in a MATLAB platform, featuring a power
distribution system with photovoltaic (PV) and wind energy conversion
Keywords: system (WECS), stabilized by a boost converter and cascaded fuzzy logic
controller (CFLC) based maximum power point tracking (MPPT) for PV and
Boost converter a PI controller for WECS. Comparative analyses demonstrate the superior
CFLC performance of the CNN classifier with an accuracy of 96.33%, outshining
CNN existing classifiers, including ANN. Furthermore, under various fault
DWT conditions, the CNN consistently achieves high accuracy, with 98% for
PV system Islanding, 95% for line-to-ground fault, and 96% for line-to-line fault. The
WECS proposed approach exhibits excellent computational efficiency, with a
training time of 10.5 hours, inference speed of 5 milliseconds, and resource
utilization of 85%, emphasizing its suitability for instantaneous fault
identification in power systems.
This is an open access article under the CC BY-SA license.

Corresponding Author:
Y. Mohamed Shuaib
Department of Electrical and Electronics Engineering
B. S. Abdur Rahman Crescent Institute of Science and Technology
Chennai, Tamilnadu, India
Email: [email protected]

1. INTRODUCTION
The growing concerns regarding global warming and environmental pollution have accentuated the
significance of RES based power production. According to International Renewable Energy Agency (IRENA),
for year 2020, the capacity of world renewable energy has expanded by more than 260 GW, which is actually
50% greater than the amount of energy generated in 2019 [1], [2]. Among the available renewable energy
sources (RES), the wind and solar energy account for a significant portion in terms of utilization rates.
However, both these RES are characterized with intermittency, which leads to the introduction of electrical
disturbances that distorts the current and voltage, resulting in several power quality disturbances (PQD). The
massive infusion of power electronic technologies, which are non-linear in nature, into the electric power
system has also resulted in the onset of serious PQD [3], [4]. These PQD has to be minimized in order to curtail
power loses and also to prevent the malfunctioning of components interfaced to power system. The power grid
faults are unpredictable and random in nature, hence the development of a precise and accurate fault detection
technique is instrumental in ensuring its safe functioning. Moreover, for maintaining balance in power system,

Journal homepage: http://ijpeds.iaescore.com

1852  ISSN: 2088-8694

a rapid fault detection model is essential as it improves the chances of quick recovery of unhealthy phases.
Thus, by restoring the unhealthy phases, the stability, PQ and transient response of the entire power system is
improved [5]-[10].
Numerous fault detection models for power distribution system are proposed so far, which includes a
communication-based protection plan in [11] that offers a back-up security in case of absence of the main
protection approach. In [12], the fault types in the output of the distributed generations are detected using abc
to dq transformation. However, the protection technique proposed in these works concentrate only on a specific
system topology or operating mode. A harmonic current based power distribution system protection plan is
proposed in [13], while a zero and positive sequence voltage magnitude and phase angle-based fault detection
approach is suggested in [14]. Due to the fluctuating fault impedances and dynamic load behaviour, it is a
challenging task to establish a suitable threshold value in the aforementioned techniques. Recently, the neural
network-based fault detection techniques are becoming increasingly popular. In these techniques, the various
types of faults are identified by extracting the line signal’s fault features as it is impossible to detect fault from
unprocessed current or voltage signal. Some of commonly used signal processing tool for extracting features
of the fault signals are Hilbert-Huang transform [15], S-transform [16] or wavelet transform [17].
For classifying the faults, machine learning techniques like support vector machines (SVM), k-nearest
neighbours (k-NN) and decision trees (DT) are utilized [18], [19]. However, the data reduction method
employed for minimizing the computational complexity, comes at the cost of information loss and lowered
accuracy. A fault discernment mechanism is envisioned on the basis of artificial neural network (ANN) is
suggested in [20], [21], while another fault discernment mechanism envisioned on the basis of probabilistic
neural network (PNN) is suggested in [22]. The process of image classification is challenging in case of ANN,
since the 2D images are required to be converted to 1D vectors, whereas the PNN are highly susceptible to
input data discrepancies. RNN is also employed in grid connected solar systems but is susceptible to stability
issues [23]. The most effective neural network for image classification is CNN and it is capable of handling
large input image database. A comprehensive examination of contemporary methodologies introduced for the
diagnosis of faults in power systems is elucidated within Table 1. The existing techniques provide valuable
contributions to fault diagnosis in power systems, but there is a notable research gap that the proposed work
aims to address. Specifically, none of the contemporary techniques discuss explicitly to integrate the
advantages of both DWT and CNN for fault identification in a power distribution system that includes RES
such as WECS and PV. This integration is precisely tailored to address the challenges posed by RES
intermittency by employing separate stabilization methods for PV and WECS.
The contributions of the proposed methodology are:
i) Integration of DWT and CNN: The proposed work introduces a novel fault detection methodology by
combining the strengths of DWT for efficient feature extraction and CN for robust classification. This
integration enhances the fault detection system's accuracy and effectiveness;
ii) Addressing challenges in RES integration: The work specifically addresses challenges associated with the
growing penetration of RES in power distribution systems. It recognizes and tackles issues such as RES
intermittency, kinetic energy insufficiency, and the lack of zero crossing current, which are prime aspects
of hybrid power system fault identification;
iii) Comprehensive power system model: The proposed framework comprises a comprehensive power
distribution system that includes both WECS and PV systems. The stability of each system is ensured
through the use of specific control methodologies, such as the boost converter and CFLC for PV and the PI
controller for WECS;
iv) Holistic evaluation of power system: The proposed work goes beyond fault detection and includes a
thorough evaluation of the power distribution system. It considers PQ issues, load variations, and influence
of faults on entire system stability. This holistic approach provides a more complete understanding of the
system's behavior under different conditions; and
v) Simulation and transparency: Simulation outcomes are presented and evaluated using MATLAB platform,
providing transparency and reproducibility. Thus, the proposed work contributes significantly to fault
diagnosis in power distribution by introducing an integrated approach, addressing challenges in RES
integration, providing a comprehensive power system model, conducting transparent simulations,
performing a comparative analysis, and demonstrating the feasibility and efficiency of the proposed
methodology.
These contributions collectively advance the understanding and application of fault detection techniques within
the framework of contemporary power systems.
The subsequent sections of the paper are meticulously organized to present a coherent and
comprehensive exploration of recommended fault identification system. Section 2 delves into materials and
methods employed in the study, providing a detailed account of the working methodology. Section 3 outlines
the intricacies of the proposed system, while the modeling intricacies of recommended system are expounded

Int J Pow Elec & Dri Syst, Vol. 15, No. 3, September 2024: 1851-1870
Int J Pow Elec & Dri Syst ISSN: 2088-8694  1853

upon in section 4 section 5 is dedicated to presenting the results and engaging in a thorough discussion.
Simulation outcomes, parameters, and waveforms are meticulously analyzed to showcase efficacy of the
envisioned fault identification approach under various fault conditions. Finally, section 6 encapsulates the
findings and insights into a succinct conclusion.

Table 1. Contemporary techniques for fault diagnosis in power systems: an elaborate analysis
Ref. Classifier/technique Feature extraction Strong points Shortcomings
[24] DT classifier Hilbert Huang and Fast and accurate fault Sensitivity to measurement quality,
empirical mode detection, classification, and DG capacity changes, and fault
decomposition location incidence angles
[25] Random forest (RF) The technique used is High accuracy, robust Sensitive to noisy data and
not explicitly mentioned generalization and effectual interpretability limitations
handing of missing data
[26] k-NN based ensemble DWT Robust feature extraction by Multiple base classifiers introduce
classifier DWT and high discrimination computational complexity
accuracy
[27] Genetic algorithm Lasso penalty Automatic fault diagnosis and Assumption of homogenous fault
based ensemble uses less training dataset characteristics and limited dataset
classifier diversity
[28] DT classifier DWT High impedance fault Performance under unpredictable
diagnosis and high noise require validation
classification accuracy
[29] Fuzzy neural network The technique used is Distributed processing, Complex and resource intensive
(FNN) not explicitly mentioned successful fault diagnosis and
selective activation
[30] Temporal constrained No feature extraction Uncertainty integration, Challenges in adapting to highly
fuzzy petri nets temporal constraints dynamic power systems
incorporation, feasibility and
matrix algorithm

2. MATERIALS AND METHODS

2.1. Dataset and simulation platform
The proposed fault detection system is evaluated using a simulated dataset generated on the MATLAB
platform. The dataset includes voltage and current signals from a hybrid power system consisting of WECS
and PV sources. These signals are recorded during normal operating conditions and under various fault
scenarios.

2.2. Data pre-processing

Prior to training the CNN, the dataset undergoes pre-processing to ensure uniformity and relevance.
This includes normalization of signal amplitudes, removal of noise, and segmentation of data into fault and
non-fault instances. In order for the network to generalize patterns from the input signals, the pre-processing
stage is essential.

2.3. CNN architecture

It is designed to efficiently capture spatial and temporal features from the fault signals. Its intricate
design executes meticulous feature extraction using convolutional layers, dimensionality mitigation using max-
pooling layers, classification using fully connected layers, and a softmax layer devoted to precise probability
estimation. The training regimen is orchestrated through the employment of Backpropagation, leveraging
stochastic gradient descent as the optimization algorithm, ensuring the network learns and refines its parameters
effectively.

2.4. DWT feature extraction

Fault signal processing involves extracting relevant features using DWT. The voltage and current
signals during fault condition onset are sensed and fed into DWT, which decomposes signals into
approximation (App) and detail (Det) coefficients. These coefficients represent large and small scale frequency
components, providing valuable information for fault classification.

2.5. CFLC MPPT

For the PV system, a boost converter with CFLC MPPT is employed to maximize power generation.
The CFLC is configured with if-then rules based on inputs such as solar irradiance, temperature, voltage,
current, and power. The cascading approach in the FLC helps reduce rule base complexity, ensuring efficient
and precise MPPT under varying operating conditions.

CNN based fault event classification and power quality enhancement in hybrid … (Abdul Quawi)
1854  ISSN: 2088-8694

2.6. Simulation parameters

Simulations are conducted with specified parameters. These parameters ensure a realistic
representation of the hybrid power system and fault scenarios. The proposed fault detection system is assessed
for its capability for precisely identifying and classifying various anomalies occurring in hybrid power system.
The CNN classifier's effectiveness, coupled with DWT feature extraction and CFLC MPPT, is analysed based
on simulation results and performance metrics such as accuracy, sensitivity, and specificity.

3. PROPOSED SYSTEM DESCRIPTION

The electrical power system in general encompasses numerous dynamic and interrelated components,
which are highly liable to electrical faults. The stability and dependability of the electrical distribution system
are negatively impacted by these failures, which can cause sizable financial losses. So, these faults are required
to be detected and cleared immediately after their occurrence. A CNN classifier based fault event detection
approach as seen in Figure 1 is envisioned in the recommended study. Moreover, the recommended fault
detection approach is applied in a hybrid power system comprising of RES such as WECS and PV. Recently,
the implementation of RES for power generation is becoming more popular on account of dwindling fossil fuel
supply and increasing environmental pollution. However, the rise in integration of RES to the power system
also causes serious protection risks in the power system. This is mainly attributable to their intermittent and
variable nature, so with the intension of stabilizing their outputs, individual control methodologies are adopted.
The PV system being the generator of low voltage supply, is coupled with a boost converter for enhancing its
output voltage. Moreover, utmost power from PV is extracted with the assistance of CFLC MPPT.
The DFIG output is meticulously conveyed to PWM rectifier, a critical component in the system
architecture. This rectifier plays a pivotal role in transforming the generated output into a stable DC form.
Facilitating this transformation is the employment of a PI controller, adding a layer of precision and control to
the stabilization process. Through a three phase VSI, the grid is linked to both of these distributed generators.
The voltage signals of the system and current are monitored constantly, to discern fault occurrence. In case of
a fault occurrence, faulty are sensed and sent as input to DWT, which in turn extracts the prominent features
of the signal. Based on these extracted features, the CNN classifier instantly detects and classifies the type of
the incurred fault. By detecting faults quickly, appropriate maintenance measures are taken to ensure power
system reliability by preventing the onset of PQDs.

Figure 1. Architecture of CNN based fault detection model in a hybrid power system

4. PROPOSED SYSTEM MODELLING

4.1. Network fault types
The various faults that occur in a power line of power distribution system are classified broadly as
series and shunt faults. The former is a result of series impedance imbalance caused due to plain break in one
or two conductors. While, latter frequently occurs in a three-phase power network during distribution of power
in the form of phase to ground (PG), phase-to-phase (PP), three PG (3 PG), and two PG (2 PG).

4.1.1. Single PG fault

This short circuit fault takes place when a specific phase line drops to ground or comes in contact to
neutral line. Moreover, it is caused due to falling of trees to the ground or as consequence of heavy rain. This
can lead to an imbalance in the system, causing a surge in current flow through the grounded phase. The faulted

Int J Pow Elec & Dri Syst, Vol. 15, No. 3, September 2024: 1851-1870
Int J Pow Elec & Dri Syst ISSN: 2088-8694  1855

phase experiences a short circuit to the ground, resulting in an increase in current and potential overheating of
equipment. Figures 2(a)-2(c) illustrates three types of single PG faults.

4.1.2. Two PG fault

This fault occurs due to the falling of two lines to the ground and it gives to fault current of higher
magnitude. The simultaneous grounding of two phases creates a more significant imbalance in the system,
leading to an increased flow of current through the faulted phases. This fault if not fixed immediately, develops
a severe three-line to-the ground fault. Various types of two PG fault with fault resistance specified as 𝑅𝑓 are
displayed in Figures 2(d)-2(f).

4.1.3. PP fault
In a three-phase system, the PP fault occurs when there is a short circuit between two specific lines.
The fault impedance magnitude of this unsymmetric fault, fluctuates over a wide range making it a challenging
task to determine its lower and upper limits. In this fault scenario, two phases of the system become directly
connected, bypassing the load impedance and creating an unintended low-impedance path for current flow
between them. This can lead to an imbalance in the system, causing changes in voltage and current magnitudes.
Figures 2(g)-2(i) illustrates the types of PP faults.

4.1.4. Three PG fault

It is a symmetric fault that often occurs and is caused because of equipment failure or falling of the
electric pole. Moreover, it generates higher amount of short circuit current and the three-phase voltage becomes
zero during this fault. During a three-phase fault, a substantial amount of short circuit current is generated
because all three phases are directly connected, bypassing the impedance of the system. This fault condition
causes a rapid increase in current magnitude, potentially leading to overheating of equipment and posing a
significant threat to the integrity of the power system. Figure 2(j) illustrates this fault.

(a) (b) (c) (d) (e) (f)

(g) (h) (i) (j)

Figure 2. Types of faults: (a) a-g, (b) b-g, (c) c-g, (d) ab-g, (e) bc-g, (f) ca-g, (g) a-b, (h) b-c, (i) a-c,
and (j) abc-g

4.2. DWT based feature extraction of fault signals

Voltage and current signals of the system during fault condition, are sensed and then provided as input
to the DWT, which subsequently analyses the signal to identify its distinct features. It aids with the
representation of signal in time domain as a collection of adjustable wavelet coefficients, which are capable of
generating wide variety of signal processing effects. By representing the signals in different domain, better
understanding of prominent features of the signal such as power spectrum, autocorrelation and periodicity is
obtained. Here, during the occurrence of fault, the distinct feature of a fault signal is represented with the
application of DWT. Moreover, for fault signal processing, a set of high pass filters (HPF) and low pass filters
(LPF) are employed, with the former handling low frequency domain signal analysis, while the latter handling
high frequency domain signal analysis. Consequently, fault signal gets disintegrated into approximation (App)

CNN based fault event classification and power quality enhancement in hybrid … (Abdul Quawi)
1856  ISSN: 2088-8694

coefficient [representation of large and small scale frequency components] and detail (Det) coefficient
[representation of small and large-scale frequency components]. DWT decomposition tree is created as a result
of App coefficient replication, as seen in Figure 3.
The occurrence of fault alters wavelet coefficients of waveforms, and these changes are also
documented for a variety of other system parameters that contain crucial fault signature data. The expression
for evaluating energy of a signal is given as (1).

𝐸𝑖 = ∑𝑛𝑗=1 𝑑𝑖𝑘
2
(1)

Where, the Det coefficient of a signal is specified as𝑑𝑖𝑘 , number of points for wavelet coefficients 𝑗 =
1,2, … , 𝑛 and scale 𝑖 = 1,2, … , 𝐼. With the occurrence of a fault or other transient events, the signal energy
changes in line with the variation in the wavelet coefficient. The required input features for presented fault
detection and classification model are obtained by evaluating the signal energy of every 3Φ signals. The DWT
applied to a signal 𝑥 at scale 𝑚 and position 𝑛 is (2).
𝑚𝑏
𝑛−𝑙𝑎
1 0 0
𝐷𝑊𝑇𝑥,𝑚,𝑛 = ∑𝑙 𝑥(𝑘)𝜓 ∗ ( ) (2)
𝑎0𝑚
√𝑎0𝑚

Where, 𝑎0𝑚 specifies scale shift parameter, 𝑙𝑎𝑚0 𝑏0 specifies the time shift parameter, and the terms 𝑚 and 𝑙 are
integer variables.

Figure 3. DWT decomposition tree

4.3. CNN based fault classification

The network fault types are classified with the aid of CNN and for the CNN to function, the fault
signals are required to be converted to images, which is accomplished with the aid of wavelet-based signal
analyser of MATLAB. The topology of the CNN as seen in the Figure 4 entails a softmax layer, max pooling,
input layer along with fully connected layers and convolutional layers. They are mainly applied for 2D image
recognition, since they are highly efficient when operated with image inputs. In addition to classification, the
different layers found in CNN are also capable of performing architecture size reduction and feature extraction.
The prominent features of the input image are extracted by convolution layer of CNN. Moreover, training of
CNN is carried out by updating a collection of filters that estimates the dot products between input and filter
values by spatially sliding across the input matrix. When these filters slide along the height and width of the
image input, a 2D activation map is generated, which at each spatial position, provides the filter response.

Figure 4. Architecture of CNN fault classifier

Int J Pow Elec & Dri Syst, Vol. 15, No. 3, September 2024: 1851-1870
Int J Pow Elec & Dri Syst ISSN: 2088-8694  1857

When the filters come across unique features in an input image like colours or edges, it gets activated
and the network is also alerted. The max-pooling layer located between the convolutional layers, reduces
overfitting by decreasing the parameters and size of the input image. The major process of classification in
CNN takes place in the fully connected layer, which comprises of several neurons that are connected to the
previous layer activation functions. The ReLU activation function applied at each convolution unit is (3).

z l , ifz l ≥ 0
f(z l ) = { (3)
0, otherwise

Where, the 𝑙 th convolutional layer output element is specified as 𝑧 𝑙 . As mentioned earlier, the
maxpooling layer, aids with the reduction of the overall computational complexity by minimizing the spatial
size of parameters and features of the input data. All the pooled features are then converted to 1D vector by the
flatten function. On the basis of features collected by the prior layers, the fully connected layer forms a new
feature map. The final layer of the CNN is the softmax layer and for 𝑀 classes, the output is given as (4).

z = [z1 , … , zM ]T = σ(h) (4)

Here, the type of fault determined in the 𝑚-𝑡ℎ category of𝑀 classes is referred as 𝑧𝑀 , the softmax
function is specified as𝜎(ℎ) and for the final fully connected layer ℎ = [ℎ1 , … , ℎ𝑀 ]𝑇 , the output is given as (5).

ehm
zm = [σ(h)]m = hj (5)
∑M
j=1 e

With the aid of training data set 𝑉, CNN parameters are developed with the intention of minimizing the loss
function. Here, the 𝑖th training sample’s loss function is the cross entropy between the target and prediction,
as in (6).

(i) (i)
Loss(z (i) ) = − ∑M
m=1 t m log (zm ) (6)

(𝑖) (𝑖)
Where, for the 𝑖th training sample, if the index of ground truth is 𝑚, then 𝑡𝑚 = 1 or otherwise𝑡𝑚 = 0. The
total training set loss is given as (7).
1
I(θ) = |V| ∑iϵV Loss(z (i) ) (7)

Where, the test set elements are specified as|.|. Moreover, in order to minimize the loss function,
Adam optimizer is selected. An image of size 224 × 224 × 3which represents the3Φ voltage or current is
provided as the CNN input. This image is accepted by the input layer, then it passes through the various filters
of the convolution layer. It then subsequently passes through ReLU and maxpooling layer. Moreover, input
data is down sampled by the pooling layer, in order to minimize overfitting. The convolution between the input
matrix and filters is the major task performed by the convolution layer and its working is greatly influenced by
the filter size. The smaller filter size leads to information loss, whereas a larger filter size leads to computational
complexity. The presented fault classification technique is implemented in a hybridized power system
encompassing wind and PV with individual control approaches, which are detailed as follows:

4.4. PV system modelling

The PV cell circuit as seen in the Figure 5 is represented as a single diode model for the ease of
understanding. The terms𝐼0 and 𝐼𝑝ℎ represents the diode saturated current and the photo generated current
respectively, while the terms 𝑅𝑃 and 𝑅𝑆 refers to the shunt and series resistance respectively. Moreover, output
current of PV cell is given as (8).
VPVcell +IPVcell RS
IPVcell = Iph − (Io (eVD ⁄αVT − 1)) − (8)
Rp

The output current of a PV module produced by combining multiple PV cells is shown as (9).

q(VPV +IPV RS ) VPV +IPV RS Ns

IPV = Iph − Io [exp ( ) − 1] − (9)
AKTNs RP N s

CNN based fault event classification and power quality enhancement in hybrid … (Abdul Quawi)
1858  ISSN: 2088-8694

Where, the terms 𝑇, 𝑞, 𝑘 𝑎𝑛𝑑 𝐴 refers to the operating temperature, electron charge, Boltzmann constant and
diode ideality factor respectively. Moreover, the number of cells that together constitute a PV module is
specified as 𝑁𝑠 .

Figure 5. Configuration of a PV module and a PV cell

4.5. Boost converter modelling

The boost converter aids with maximizing the solar power generation 𝑃𝑃𝑉 by varying output
voltage 𝑉𝑃𝑉 of PV array. The input voltage to converter is 𝑉𝑃𝑉 , 𝑖𝐿 represents the inductor current and 𝑉𝑜𝑢𝑡
indicates the voltage output. Input capacitor which is shunt coupled to the PV generator is eliminated as seen
in Figure 6, for simplifying the designing process of the MPPT controller.
On considering the continuous conduction, the boost converter operates in two switching modes,
which are:
- On-state dynamics: In case of switch ON condition, the dynamics of output voltage and input current is
expressed as (10).

dIPV VPV
=
{dV dt L
−Vout
(10)
out
=
dt RC

- Off-state dynamics: In case of switch OFF condition, the boost converter’s state space equation is given
as (11).

dIPV VPV −Vout

=
{dVdt IPV
L
Vout
(11)
out
= −
dt C RC

A time invariant nonlinear system is represented generally as (12).

Ẋ = f(X) + g(X)u (12)

Where, 𝑢 ∈ [0 1] and the vector state space is represented as 𝑋 = [𝑥1 𝑥2 … 𝑥𝑛 ]𝑇 . On the basis of (12), the model
of boost converter is formulated by multiplying the duty cycle D with (10) and (1-D) with (11). Moreover, the
state-space average approach is also used, in which (13) and (14).

Dϵ[0 1] (13)

X = [IPV Vout ]T (14)

Int J Pow Elec & Dri Syst, Vol. 15, No. 3, September 2024: 1851-1870
Int J Pow Elec & Dri Syst ISSN: 2088-8694  1859

In continuous mode, the boost converter model is given as (15).

dIPV VPV Vout

= − (1 − D)
{dVdt L
−Vout
L
IPV
(15)
out
= + (1 − D)
dt RC C

The extracted power from the PV is further maximized with the aid of an appropriate MPPT approach. Here, a
CFLC is utilized as MPPT controller, which adjusts D of boost converter to allow extraction of maximum power.

Figure 6. Boost converter without input capacitor

4.6. CFLC MPPT technique

The fuzzy logic controller (FLC) is an effective mathematical tool capable of handling vague, unclear,
uncertain and imprecise data. The mapping of the input to the output is carried out with the aid of if-then rules
in an FLC and the term rule base refers to the collection of these if-then rules. The complexity of the rule base
has a significant impact on the FLC's performance because a more complicated rule base necessitates a longer
execution time. Thereby, with the aim of reducing the rule base complexity, a cascading approach is proposed
for FLC in this work. The cascading technique is implemented by connecting several FLCs in series, with the
output from one FLC serving as input to the following FLC. The CFLC is used as an MPPT controller and the
inputs to the MPPT are solar irradiance, temperature,𝑉𝑃𝑉 , 𝐼𝑃𝑉 and 𝑃𝑃𝑉 . For a normal FLC with 𝑛 membership
functions and 𝑚 inputs, the total number of rules is expressed as (16).

Total number of rules = mn (16)

As mentioned above, five inputs are considered for the MPPT controller. Thus, it is resulting in the
requirement of 55 = 3125 rules, in case of a normal FLC with five membership functions. However, in case
of CFLC MPPT as seen in Figure 7, requires a total of 52 + 52 + 52 + 52 = 100 rules, thus successfully
reducing the computational complexity of the MPPT controller.

Figure 7. Structure of CFLC MPPT

The FLC being an intelligent technique, is tuned on the basis of human experience and knowledge
about the working of the system. Moreover, compared to conventional MPPT techniques, it overcomes the
CNN based fault event classification and power quality enhancement in hybrid … (Abdul Quawi)
1860  ISSN: 2088-8694

problem of overshoot and is capable of maintaining the operating point at MPP in spite of fluctuations in
temperature and irradiance. The inputs of the FLC are mapped to the fuzzy variable in the fuzzification block.
The actual expert data are presented in the inference block through the membership rules contained in if-then
statements. The linguistic inference rules are then converted to crip numerical values in the output
defuzzification block. In case of CFLC MPPT, the change in duty ratio command for the boost converter is
attained as output.

4.7. Modelling of DFIG based WECS

The characteristics of a wind turbine, which generates mechanical energy through the conversion of
kinetic energy of the wind is given as (17)-(19).
1
λi = (17)
1⁄λ+0.08β−0.035/(β3 +1)

116
Cp (λ, β) = 0.645( − 0.4β − 5)e−21⁄λi (18)
λi

0.5ρπR2 Cp (λ,β)V3
wind
Tm = (19)
ωWT

Where, the terms 𝜌, 𝛽, 𝜆, 𝐶𝑝 𝑎𝑛𝑑 𝑅 refers to the air density, pitch angle, tip speed ratio, power coefficient and
blade radius respectively. A highly efficient AC asynchronous variable-speed generator with decreased
mechanical stress, greater energy harvesting, and improved reactive power controllability is known as the
DFIG. Figures 8(a) and (b) illustrate the structure and equivalent circuit of a DFIG. The DFIG is modelled with
the aid of subsequent expressions. The stator voltage is estimated using (20).
̅̅̅̅s
1 dΨ 1 dθs
v̅s = −R s i̅s + +j ̅̅̅
Ψs (20)
ωb dt ωb dt

The rotor voltage is expressed as (21), the stator flux linkage is expressed using (22), while the
electromagnetic torque is expressed as (23).
̅̅̅̅r
1 dΨ 1 dθr
v̅r = −R r i̅r + +j ̅̅̅
Ψr (21)
ωb dt ωb dt

̅̅̅
Ψs = Ls (−i̅s + i̅), ̅̅̅ ̅̅̅
r Ψr = Ψs + Lkr i̅r (22)

̅̅̅s ir ) , ps = −Re(v̅s is )
t e = −Im (Ψ (23)

Where the stator voltage, current, inductance and resistance are specified as 𝑣𝑠 , 𝑖𝑠 , 𝐿𝑠 𝑎𝑛𝑑 𝑅𝑠 respectively,
while the rotor voltage, current, inductance and resistance are specified as𝑣𝑟 , 𝑖𝑟 , 𝐿𝑟 𝑎𝑛𝑑 𝑅𝑟 respectively. The
positive torque is specified as𝑡𝑒 , while the rotor and stator flux are designated as Ψ𝑟 and Ψ𝑠 respectively. To
obtain a DC voltage output, a PWM rectifier is linked to the DFIG-based WECS's output. The PWM rectifier's
output is stabilized using the PI controller, which provides the appropriate error compensation.

(a) (b)

Figure 8. DFIG topology: (a) DFIG structure (b) DFIG equivalent circuit

5. RESULTS AND DISCUSSION

The process of fault detection using CNN classifier and DWT is proposed. The proposed fault detection
scheme is considered in hybrid power system comprising of RES such as WECS and PV. Since both these RES

Int J Pow Elec & Dri Syst, Vol. 15, No. 3, September 2024: 1851-1870
Int J Pow Elec & Dri Syst ISSN: 2088-8694  1861

are characterized with intermittency, individual control methodologies are employed for stabilizing their outputs.
Through accurate detection and correction of faults, the PQ of the entire system is enhanced. Moreover, by
taking suitable maintenance measures, the breakdown of the entire power distribution system is also prevented.
The simulation setup of suggested work is presented in Figure 9. The system includes a combination of WECS
and a PV system, where the PV system incorporates a boost converter and a closed-loop CFLC MPPT for
optimal power extraction. The output from the DFIG undergoes transformation through a PWM rectifier,
controlled by a PI controller. Both the PV system and DFIG are linked to grid via a three-phase VSI. Continuous
monitoring of voltage and current signals enables prompt fault detection. In the event of a fault, the signals are
processed through DWT and the extracted features are input into a CNN classifier for rapid and accurate fault
classification. The processing of signals is added as sub blocks in the simulation setup. This setup aims to
expedite fault identification, allowing for timely maintenance actions to prevent PQDs and ultimately ensuring
power system reliability, particularly in the context of intermittent and variable RES integration.

Figure 9. Simulation setup of the proposed work

5.1. Simulation results

The overall fault detection system is analysed on the basis of the simulation outcome derived from
MATLAB. A simulation model for the suggested DWT-CNN based fault identification approach is developed
and tested for its effectiveness in maintaining PQ in a power distribution system. Table 2 also includes a
tabulation of the parameters for the developed simulation model.
The PV system unlike the conventional fossil fuel-based power systems, delivers a variable voltage
supply as output. The PV panel output voltage waveform given in Figure 10(a), indicates that a rise in voltage
level from 185 V to 195 V is observed at 0.2 s. This variation in voltage level is attributable to the operating
condition changes such as variation in ambient temperature or solar irradiance. However, on the basis of the
boost converter output voltage waveform shown in Figure 10(b), it is observed to maintain a 600 V at 0.2 s
regardless of the input voltage fluctuation. This is mainly due to the effectiveness of the CFLC MPPT and the
boost converter.
The waveform representing the DFIG voltage output is given in Figure 11(a). It is evident that the
voltage output ranges at +600 V to -600 V and this AC voltage is then supplied to the PWM rectifier. The
PWM rectifier is effective in delivering a stabilized DC voltage of 600 V as output from 0.15 s, with the
assistance of PI controller as shown in Figure 11(b). The outputs showcasing the turbine rotor speed and torque
are provided in Figure 12. The three currents of the three phases of the stator are provided in Figure 13.
DC-link voltage is noted to be a stable 600V from 0.075s after initial variations as seen in
Figure 14. In a power distribution system, the abrupt addition or removal of loads induces PQDs such as voltage
swell and sag as displayed in Figure 15. Voltage swell issue is characterized with instantaneous increase in
voltage level, whereas the issue of voltage sag is characterized with instantaneous decrease in voltage level.
Both these PQDs are required to be detected and compensated swiftly for ensuring the stable power distribution
system working.
Here, PI controller and dq theory aids with the achievement of grid voltage synchronization and a
stable voltage and current of range -400 V to +400 V and -13.5 A to +13.5 A respectively are maintained as
seen in Figure 16(a)-(b). For better understanding, the phase 𝐴 waveform of both grid voltage and current are
considered as illustrated in Figure 16(c).

CNN based fault event classification and power quality enhancement in hybrid … (Abdul Quawi)
1862  ISSN: 2088-8694

Table 2. Parameter specifications

Parameters Values Parameters Values
No. of series cells 36 Operating voltage and current 16.8 V and 5.8 A
No. of panels 10 Switching frequency 1 kHz
DC-link voltage 600-1500 V Load type Variable
Area of each cell 125×31.25 mm

(a) (b)

Figure 10. Output voltage waveforms: (a) PV panel and (b) boost converter

(a) (b)

Figure 11. WECS waveforms representing: (a) DFIG output and (b) PWM rectifier

(a) (b)

Figure 12. Waveforms showcasing wind turbine: (a) rotor speed and (b) torque

Int J Pow Elec & Dri Syst, Vol. 15, No. 3, September 2024: 1851-1870
Int J Pow Elec & Dri Syst ISSN: 2088-8694  1863

(a) (b)

(c)

Figure 13. Output Waveforms of DFIG current: (a) phase A current, (b) DFIG phase B current,
and (c) DFIG phase C current

Figure 14. DC link voltage

(a) (b)

Figure 15. Source voltage affected by PQ issues of (a) voltage swell and (b) voltage sag
CNN based fault event classification and power quality enhancement in hybrid … (Abdul Quawi)
1864  ISSN: 2088-8694

(a) (b)

(c)

Figure 16. Grid side waveforms representing (a) voltage, (b) current, and (c) voltage and current waveform

Magnitude of real power as observed from Figure 17(a) is 8500 W, while the reactive power is noted
to be below -150 VAR as seen in Figure 17(b). Since, both voltage and current waveforms are in phase, a
power factor value of one is observed in Figure 17(c). The power factor value of one also implies that the
amount of harmonic content in the grid side is very low. This is confirmed with the assistance of the THD
waveform given in Figure 18, which shows that the value of THD is 0.89%.

(a) (b)

(c)

Figure 17. Grid side: (a) real power waveform, (b) reactive power waveform, and (c) power factor waveform

Int J Pow Elec & Dri Syst, Vol. 15, No. 3, September 2024: 1851-1870
Int J Pow Elec & Dri Syst ISSN: 2088-8694  1865

Figure 18. THD waveform

5.2. Fault analysis

The proposed fault detection system comprising of DWT and CNN classifier, undergoes a
comprehensive evaluation under conditions simulating half the normal load to assess its fault detection
capabilities across a variety of fault scenarios. This methodical analysis aims to validate the robustness and
sensitivity of the system when the electrical system is not operating at full capacity. It is crucial for
understanding its performance under less-than-ideal conditions.

5.2.1. Islanding condition

Islanding is an unsafe and critical condition during which the RES based distributed generators
continues to supply power even when there is no power from the grid. The problem of islanding is dangerous
to the utility workers and it causes damage to the customer appliances and inverters. Hence, the earliest
detection of islanding is considered important.
The CNN classifier detects the islanding condition on the basis of grid current and voltage waveforms
given in Figure 19. An abrupt variation in grid voltage is noted during the islanding condition that occurs
between 0.5 to 0.6 seconds. The grid current remains unchanged, however a corresponding variation in real
power and reactive power is observed during this time interval.

5.2.2. Line-to-ground fault

This fault typically arises from incidents such as a tree collapsing onto the power line, especially
during periods of heavy rainfall. This fault condition materializes when the power line makes contact with the
ground or comes into connection with the neutral line due to external factors like adverse weather conditions.
Specifically, if the line physically descends to the ground or establishes contact with the neutral line, it results
in what is referred to as a line-to-ground fault. This fault scenario can be triggered by a variety of environmental
factors, with the consequences analyzed based on the extent and nature of the fault occurrence.
CNN classifier detects this fault on basis of current and voltage waveform displayed in Figure 20. The
Line-to-Ground fault is observed between 0.5 to 0.6 seconds, during which the grid voltage witnesses a slight
dip in its magnitude, while the grid current undergoes rapid distortions. Similarly, the active power also
decreases in its value, whereas the reactive power experiences fluctuations during the occurrence of this fault.

5.2.3. Line-to-line fault

In a triphasic system, an asymmetrical fault arises when a short circuit arises between lines, termed as
line-to-line fault. This type of fault manifests when an unintended connection or disruption arises between two
distinct phases within the power distribution network. The line-to-line fault is characterized by a disruption in
the equilibrium of the electrical system, leading to distinctive alterations in grid voltage and current. This fault
condition occurrence can be attributed to various factors, such as equipment malfunction or external
disturbances. The unsymmetrical nature of the fault introduces complexities in the electrical parameters,
requiring a meticulous analysis for accurate detection and subsequent corrective measures. The recognition of
line-to-line faults is pivotal in maintaining the integrity of the power system, as prompt identification enables
swift intervention to prevent cascading failures and minimize potential damage to connected equipment.
The line-to-line fault is detected between 0.5 to 0.6 seconds by the CNN classifier. As seen in
Figure 21, grid voltage is almost zero and grid current is affected by heavy distortions during this fault
condition. The reactive power experiences a sudden increase, while the active power drops to zero. The CNN
CNN based fault event classification and power quality enhancement in hybrid … (Abdul Quawi)
1866  ISSN: 2088-8694

classifier's performance is analyzed on the basis of accuracy, sensitivity and specificity in Figure 22(a) and it
is observed that the CNN classifier outshines the functioning of all the other existing classifiers in these
categories. It offers an excellent accuracy of 96.33% with sensitivity and specificity of 96%, which is
comparatively better than the performance of ANN classifier as seen in Table 3. Moreover, the efficiency of
the proposed CFLC MPPT is 91% as given in Figure 22(b). Table 4 presents a comprehensive comparison of
the performance of various fault detection systems under diverse fault conditions. The proposed CNN exhibits
an overall accuracy of 96.33%, which is significantly greater than the average accuracy of other models
(ranging from 84% to 92%). This emphasizes the comprehensive effectiveness of the CNN-based approach
across different fault conditions.

Figure 19. Islanding condition waveforms

Figure 20. Parameters noted in case of line-to-ground fault

Int J Pow Elec & Dri Syst, Vol. 15, No. 3, September 2024: 1851-1870
Int J Pow Elec & Dri Syst ISSN: 2088-8694  1867

Figure 21. Parameters noted in case of line-to-line fault

Table 5 offers a comparative overview of the classifiers based on their training time, inference speed,
and resource utilization, providing insights into the efficiency and computational demands of each model. The
proposed CNN exhibits a training time of 10.5 hours, an inference speed of 5 milliseconds, and a resource
utilization of 85%. These metrics collectively demonstrate the efficiency and computational effectiveness of
the proposed approach in training the model, making rapid predictions during inference, and utilizing system
resources optimally. In comparison, the DT classifier demonstrates a shorter training time of 2.3 hours but with
a slightly higher inference speed of 15 milliseconds and a resource utilization of 70%. SVM, RF, and FNN
show varying performance across these metrics, while the ANN classifier requires the longest training time
(20.1 hours) with moderate inference speed (12 milliseconds) and resource utilization (60%). These results
provide valuable insights into the trade-offs between training time, inference speed, and resource utilization
for each classifier. The proposed CNN stands out for achieving a balanced performance across these metrics,
showcasing its suitability for real-time fault detection applications in power systems.

(a) (b)

Figure 22. Comparison charts for (a) classifiers and (b) efficiency

Table 3. Comparison of CNN against ANN classifier

Classifiers Accuracy Sensitivity Specificity
ANN 87% 95% 93%
CNN 96.33% 96% 96%

CNN based fault event classification and power quality enhancement in hybrid … (Abdul Quawi)
1868  ISSN: 2088-8694

Table 4. Fault detection system performance comparison under various fault conditions
Fault condition CNN DT SVM RF FNN ANN
Islanding 98% 90% 92% 93% 94% 91%
Line-to-ground fault 95% 87% 88% 89% 92% 86%
Line-to-line fault 96% 85% 86% 87% 90% 84%
Overall accuracy 96.33% 87.33% 88.67% 89.67% 92% 87%

Table 5. Computational efficiency comparison of classifiers

Classifier Training time (hrs) Inference speed (ms) Resource utilization
CNN 10.5 5 85
DT 2.3 15 70
SVM 8 8 80
RF 12.2 7 75
FNN 15.8 10 65
ANN 20.1 12 60

5.3. Discussions
The proposed fault detection system, integrating DWT and CNN, demonstrates notable effectiveness
in enhancing the PQ and reliability of a hybrid power system comprising WECS and PV. The simulation
results, conducted in MATLAB, provide insights into the system's performance under various conditions. The
stability of both WECS and PV, achieved through individual control methodologies, ensures the system's
resilience against faults and contributes to maintaining a stable PQ. The PV system, characterized by variable
voltage output, exhibits stable voltage maintenance through the boost converter and CFLC based MPPT.
Similarly, the WECS output is stabilized through a PI controller. The fault analysis, conducted under different
conditions such as islanding, line-to-ground fault, and line-to-line fault, showcases the CNN classifier's robust
performance, with an overall accuracy of 96.33%. The comparative analysis against other classifiers, including
ANN, highlights the superiority of the proposed CNN classifier in terms of accuracy, sensitivity, and
specificity. Additionally, the computational efficiency comparison reveals that the CNN model achieves a
balanced performance with a training time of 10.5 hours, inference speed of 5 milliseconds, and resource
utilization of 85%. The implications of these findings extend to the future development of fault detection
systems in power distribution. The robustness demonstrated by the CNN classifier, coupled with efficient
computational performance, positions it as a promising tool for real-time fault detection applications. This
research contributes valuable insights into advancing power system stability and reliability in the context of
increasing renewable energy source integration.

6. CONCLUSION
The increased penetration of RES in power system has resulted in the existence of severe protection
issues, PQDs and faults. Therefore, to improve stability, PQ, and reliability of power system, these faults are
required to be identified instantly without delay. So, a CNN classifier-based fault detection is proposed in this
work for identifying faults taking place in a hybrid power system consist of wind and PV. Moreover, features
of the fault signal are extracted with the aid of DWT. Furthermore, individual control approaches are applied
for wind and PV for stabilizing their outputs. Consequently, a boost converter along with CFLC MPPT is
employed for PV, while PI controller is employed for WECS. The MATLAB platform is utilized for carrying
out the simulation of the proposed fault detection approach. Moreover, the classification capability of CNN is
tested for various kinds of fault such as line-to-line fault, islanding and line-to-ground faults. On the basis of
the obtained outcomes, the CNN classifier offers an exceptional classification accuracy of 96.33% and it
outperforms other available classifiers such as ANN, SVM, and RF.

REFERENCES
[1] M. K. H. Rabaia et al., “Environmental impacts of solar energy systems: A review,” Science of The Total Environment, vol. 754, p.
141989, Feb. 2021, doi: 10.1016/j.scitotenv.2020.141989.
[2] R. M. Elavarasan et al., “A Comprehensive Review on Renewable Energy Development, Challenges, and Policies of Leading Indian
States with an International Perspective,” IEEE Access, vol. 8, pp. 74432–74457, 2020, doi: 10.1109/ACCESS.2020.2988011.
[3] C. I. Garcia, F. Grasso, A. Luchetta, M. C. Piccirilli, L. Paolucci, and G. Talluri, “A Comparison of Power Quality Disturbance
Detection and Classification Methods Using CNN, LSTM and CNN-LSTM,” Applied Sciences, vol. 10, no. 19, p. 6755, Sep. 2020,
doi: 10.3390/app10196755.
[4] V. Babu, K. S. Ahmed, Y. M. Shuaib, and M. Mani, “A novel intrinsic space vector transformation based solar fed dynamic voltage
restorer for power quality improvement in distribution system,” Journal of Ambient Intelligence and Humanized Computing, 2021,
doi: 10.1007/s12652-020-02831-0.

Int J Pow Elec & Dri Syst, Vol. 15, No. 3, September 2024: 1851-1870
Int J Pow Elec & Dri Syst ISSN: 2088-8694  1869

[5] R. Eslami, S. H. H. Sadeghi, H. Askarian-Abyaneh, and A. Nasiri, “A Novel Method for Fault Detection in Future Renewable
Electric Energy Delivery and Management Microgrids, Considering Uncertainties in Network Topology,” Electric Power
Components and Systems, vol. 45, no. 10, pp. 1118–1129, 2017, doi: 10.1080/15325008.2017.1292433.
[6] Y. Zhang and W. Wei, “Decentralised coordination control strategy of the PV generator, storage battery and hydrogen production
unit in islanded AC microgrid,” IET Renewable Power Generation, vol. 14, no. 6, pp. 1053–1062, 2020, doi: 10.1049/iet-
rpg.2019.0842.
[7] G. Magdy, G. Shabib, A. A. Elbaset, and Y. Mitani, “A novel coordination scheme of virtual inertia control and digital protection
for microgrid dynamic security considering high renewable energy penetration,” IET Renewable Power Generation, vol. 13, no. 3,
pp. 462–474, 2019, doi: 10.1049/iet-rpg.2018.5513.
[8] A. Prasad, J. Belwin Edward, and K. Ravi, “A review on fault classification methodologies in power transmission systems: Part—
I,” Journal of Electrical Systems and Information Technology, vol. 5, no. 1, pp. 48–60, 2018, doi: 10.1016/j.jesit.2017.01.004.
[9] S. G. Reddy, S. Ganapathy, and M. Manikandan, “Three Phase Four Switch Inverter Based DVR for Power Quality Improvement
with Optimized CSA Approach,” IEEE Access, vol. 10, pp. 72263–72278, 2022, doi: 10.1109/ACCESS.2022.3188629.
[10] V. Babu, K. Shafeeque Ahmed, Y. Mohamed Shuaib, and M. Manikandan, “Power Quality Enhancement Using Dynamic Voltage
Restorer (DVR)-Based Predictive Space Vector Transformation (PSVT) with Proportional Resonant (PR)-Controller,” IEEE
Access, vol. 9, pp. 155380–155392, 2021, doi: 10.1109/ACCESS.2021.3129096.
[11] E. Sortomme, S. S. Venkata, and J. Mitra, “Microgrid protection using communication-assisted digital relays,” IEEE Transactions
on Power Delivery, vol. 25, no. 4, pp. 2789–2796, 2010, doi: 10.1109/TPWRD.2009.2035810.
[12] H. Al-Nasseri, M. A. Redfern, and F. Li, “A voltage based protection for micro-grids containing power electronic converters,” 2006
IEEE Power Engineering Society General Meeting, PES, 2006, doi: 10.1109/pes.2006.1709423.
[13] H. Al-Nasseri and M. A. Redfern, “Harmonics content based protection scheme for micro-grids dominated by solid state
converters,” 2008 12th International Middle East Power System Conference, MEPCON 2008, pp. 50–56, 2008, doi:
10.1109/MEPCON.2008.4562361.
[14] A. Hooshyar, E. F. El-Saadany, and M. Sanaye-Pasand, “Fault Type Classification in Microgrids Including Photovoltaic DGs,”
IEEE Transactions on Smart Grid, vol. 7, no. 5, pp. 2218–2229, 2016, doi: 10.1109/TSG.2015.2451675.
[15] A. Gururani, S. R. Mohanty, and J. C. Mohanta, “Microgrid protection using Hilbert-Huang transform based-differential scheme,”
IET Generation, Transmission and Distribution, vol. 10, no. 15, pp. 3707–3716, 2016, doi: 10.1049/iet-gtd.2015.1563.
[16] T. Gush, S. B. A. Bukhari, K. K. Mehmood, S. Admasie, J. S. Kim, and C. H. Kim, “Intelligent fault classification and location
identification method for microgrids using discrete orthonormal stockwell transform-based optimized multi-kernel extreme learning
machine,” Energies, vol. 12, no. 23, 2019, doi: 10.3390/en12234504.
[17] Y. Y. Hong, Y. H. Wei, Y. R. Chang, Y. Der Lee, and P. W. Liu, “Fault detection and location by static switches in microgrids
using wavelet transform and adaptive network-based fuzzy inference system,” Energies, vol. 7, no. 4, pp. 2658–2675, 2014, doi:
10.3390/en7042658.
[18] V. Le, X. Yao, C. Miller, and B. H. Tsao, “Series DC Arc Fault Detection Based on Ensemble Machine Learning,” IEEE
Transactions on Power Electronics, vol. 35, no. 8, pp. 7826–7839, 2020, doi: 10.1109/TPEL.2020.2969561.
[19] F. Ye, Z. Zhang, K. Chakrabarty, and X. Gu, “Board-level functional fault diagnosis using multikernel support vector machines and
incremental learning,” IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems, vol. 33, no. 2, pp. 279–
290, 2014, doi: 10.1109/TCAD.2013.2287184.
[20] Y. Y. Hong and M. T. A. M. Cabatac, “Fault Detection, Classification, and Location by Static Switch in Microgrids Using Wavelet
Transform and Taguchi-Based Artificial Neural Network,” IEEE Systems Journal, vol. 14, no. 2, pp. 2725–2735, 2020, doi:
10.1109/JSYST.2019.2925594.
[21] B. Padakanti, I. A. Chidambaram, and M. Mani, “An ANN Based MPPT for Power Monitoring in Smart Grid using Interleaved
Boost Converter,” Tehnicki Vjesnik, vol. 30, no. 2, pp. 381–389, 2023, doi: 10.17559/TV-20220820194302.
[22] S. Mishra, C. N. Bhende, and B. K. Panigrahi, “Detection and Classification of Power Quality Disturbances Using S-Transform and
Probabilistic Neural Network,” IEEE Transactions on Power Delivery, vol. 23, no. 1, pp. 280–287, Jan. 2008, doi:
10.1109/TPWRD.2007.911125.
[23] T. P. Kumar, S. Ganapathy, and M. Manikandan, “Improvement of voltage stability for grid connected solar photovoltaic systems
using static synchronous compensator with recurrent neural network,” Electrical Engineering and Electromechanics, vol. 2022, no.
2, pp. 69–77, 2022, doi: 10.20998/2074-272X.2022.2.10.
[24] M. Shaik, A. G. Shaik, and S. K. Yadav, “Hilbert–Huang transform and decision tree based islanding and fault recognition in
renewable energy penetrated distribution system,” Sustainable Energy, Grids and Networks, vol. 30, 2022, doi:
10.1016/j.segan.2022.100606.
[25] Z. El Mrabet, N. Sugunaraj, P. Ranganathan, and S. Abhyankar, “Random Forest Regressor-Based Approach for Detecting Fault
Location and Duration in Power Systems,” Sensors, vol. 22, no. 2, 2022, doi: 10.3390/s22020458.
[26] K. S. V. Swarna, A. Vinayagam, M. Belsam Jeba Ananth, P. Venkatesh Kumar, V. Veerasamy, and P. Radhakrishnan, “A KNN
based random subspace ensemble classifier for detection and discrimination of high impedance fault in PV integrated power
network,” Measurement: Journal of the International Measurement Confederation, vol. 187, 2022, doi:
10.1016/j.measurement.2021.110333.
[27] A. Eskandari, M. Aghaei, J. Milimonfared, and A. Nedaei, “A weighted ensemble learning-based autonomous fault diagnosis
method for photovoltaic systems using genetic algorithm,” International Journal of Electrical Power and Energy Systems, vol. 144,
2023, doi: 10.1016/j.ijepes.2022.108591.
[28] T. Biswal and S. K. Parida, “A novel high impedance fault detection in the micro-grid system by the summation of accumulated
difference of residual voltage method and fault event classification using discrete wavelet transforms and a decision tree approach,”
Electric Power Systems Research, vol. 209, 2022, doi: 10.1016/j.epsr.2022.108042.
[29] C. Li et al., “Takagi–Sugeno fuzzy based power system fault section diagnosis models via genetic learning adaptive GSK
algorithm,” Knowledge-Based Systems, vol. 255, 2022, doi: 10.1016/j.knosys.2022.109773.
[30] B. Xu, X. Yin, X. Yin, Y. Wang, and S. Pang, “Fault diagnosis of power systems based on temporal constrained fuzzy petri nets,”
IEEE Access, vol. 7, pp. 101895–101904, 2019, doi: 10.1109/ACCESS.2019.2930545.

CNN based fault event classification and power quality enhancement in hybrid … (Abdul Quawi)
1870  ISSN: 2088-8694

BIOGRAPHIES OF AUTHORS

Abdul Quawi graduated in B. Tech in Electrical and Electronics Engineering,

M.Tech. in Electrical Power Engineering and pursuing Ph.D. Power Quality at B. S. Abdur
Rahman Crescent Institute of Science and Technology-A deemed to be University, Chennai,
Tamil Nadu, India. Presently, he is working as an electrical instructor in Maulana Azad
National Urdu University Hyderabad, Telangana, India. He worked as assistant professor in
Alhabeeb College of Engineering & Technology Hyderabad, Telangana, India from Janurary
2013 to February 2015. He is having more than 8 years of teaching experience in the
department of Electrical and Electronics Engineering. He can be contacted at email:
[email protected].

Y. Mohamed Shuaib graduated in B.E in Electrical and Electronics Engineering,

M.E in Power Systems Engineering and Ph.D. in Electrical Engineering at Jawaharlal Nehru
Technological University, Telengana, India in November, 2015. Presently, he is working as
professor of the Department of Electrical and Electronics Engineering at B. S. Abdur Rahman
Crescent Institute of Science and Technology-A deemed to be University, Chennai,
Tamilnadu, India. He is having more than 20 years of teaching experience in the Department
of Electrical and Electronics Engineering. His research area is power quality, power system.
He can be contacted at email: [email protected].

M. Manikandan graduated in B.E in Electrical, M.E in Power Electronics and

Drives. He was awarded Ph.D. in EEE at Anna University, Chennai. He is a member of IEI,
ISTE and IAENG. He has published 17 papers in international and national level journals.
Presently, he is working as professor of the Department of Electrical and Electronics
Engineering at Jyothishmathi Institute of Technology and Science Karimnager, Telangana,
India. He is having more than 17 years of teaching experience in the Department of Electrical
and Electronics Engineering. His research area is power quality, power system. He can be
contacted at email: [email protected].

Int J Pow Elec & Dri Syst, Vol. 15, No. 3, September 2024: 1851-1870