INTERNATIONAL JOURNAL of RENEWABLE ENERGY RESEARCH

R. Venkatasubramanian et al., Vol.14, No.2, June, 2024

Integrating AI-driven Fault Detection and Protection

Technique for Electric Power Components and
Systems
R. Venkatasubramanian* , G. Diwakar** , P.Subhashini*** , V. Venkata Kumar**** , K. Rayudu***** ,
J. Samson lsaac****** , K. Bhanu Teja******* , Dr. A. Rajaram********
*
Department of Electrical and Electronics Engineering, New Prince Shri Bhavani College of Engineering and Technology,
Chennai, Tamil Nadu 600073, India
**
Department of Mechanical Engineering, Koneru Lakshmaiah Education Foundation, Vaddeswaram, Andhra Pradesh 522302,
India
***
Department of Information Technology, Vel Tech Multi Tech Dr.Rangarajan Dr.Sakunthala Engineering College, Chennai,
Tamil Nadu 600062, India
****
Department of Mathematics, Aditya Engineering College(A), Surampalem, Andhra Pradesh 533437, India
*****
Department of Electrical and Electronics Engineering, B. V. Raju Institute of Technology, Narsapur, Telangana 502313,
India.
******
Department of Biomedical Engineering, Karunya lnstitute of Technology and Sciences, Coimbatore, Tamil Nadu 641114,
India
*******
Department of Electrical and Electronics Engineering, Loyola Institute of Technology, Mevalurkuppam ‘B’ Village,
Chennai, Tamil Nadu 600123, India
********
Department of Electronics and Communication Engineering, E.G.S Pillay Engineering
College, 611 002 Nagapattinam, India
([email protected], [email protected], [email protected], [email protected], [email protected],
[email protected], [email protected], [email protected])

*Corresponding Author; R. Venkatasubramanian, Department of Electrical and Electronics Engineering, New Prince Shri
Bhavani College of Engineering and Technology, Chennai, Tamil Nadu 600073, India, [email protected]

Received: 09.08.2023 Accepted:11.12.2023

Abstract: Electric power systems are critical infrastructures that require continuous monitoring and protection to ensure reliable
operation. Fault detection and protection play a crucial role in maintaining the stability and integrity of power components and
systems. This research paper presents a comprehensive approach to enhancing fault detection and protection techniques in electric
power systems by integrating Artificial Intelligence (AI). The proposed model leverages AI-driven techniques, including Deep
Forest, Support Vector Machines (SVM), and Neural Networks (NN), for effective fault detection and protection. Deep Forest
serves as a feature extractor, capturing informative representations of fault data, while SVM and NN classifiers ensure accurate
fault type classification and decision-making. Extensive experiments and evaluations demonstrate the hybrid model's superior
performance, achieving 98.57% accuracy and highlighting its potential to advance fault detection and protection in electric power
systems.

Keywords: Artificial Intelligence, Deep Forest, Support Vector Machines (SVM), Neural Networks (NN), Fault Detection and
Protection.

1. Introduction electrical infrastructure. This research delves into advanced fault

detection and protection techniques driven by Artificial
Electric power systems are the backbone of modern Intelligence (AI) to enhance the reliability of electric power
civilization, providing a steady supply of electricity for various systems.
applications. Ensuring the reliability and stability of these
systems is critical to avoid disruptions, blackouts, and potential Integrating AI-driven techniques into fault detection and
damages. By quickly detecting and reducing defects, fault protection processes offers several advantages. SVM, Deep
detection and protection play a crucial part in protecting the Forest, and Neural Networks (NN) are examples of AI
algorithms that excel in pattern recognition and can precisely
 INTERNATIONAL JOURNAL of RENEWABLE ENERGY RESEARCH
R. Venkatasubramanian et al., Vol.14, No.2, June, 2024
identify complex failure patterns in power systems. This model aims to enable real-time monitoring and swift fault
accuracy leads to swift and precise fault detection. AI-based detection, ensuring timely mitigation actions to prevent further
systems can continuously monitor power system parameters and system instability or damage. The model aspires to possess
perform real-time fault detection [1]. The ability to detect faults adaptive learning capabilities, allowing it to continuously learn
as they occur enables swift response and timely mitigation, from new fault data and adapt to changing power system
minimizing downtime and potential damage. conditions, improving its performance over time.

AI models possess the capability to adapt to changing power The proposed AI-driven fault detection and protection
system conditions and evolving fault patterns. With continuous model aims to contribute significantly to enhancing the
learning and improvement, AI-driven fault detection systems reliability of electric power systems. Its key contributions
become more adept at recognizing and handling diverse fault include:
scenarios. By detecting faults early on, AI-driven systems
facilitate proactive maintenance and repair, reducing downtime ➢ By accurately detecting faults and initiating timely
and associated costs. This proactive approach enhances the protective measures, the model seeks to enhance the overall
overall reliability and availability of the electric power system. reliability and stability of electric power systems, reducing the
risk of widespread blackouts.
Despite the advantages, AI-driven techniques also have some ➢ Through early fault detection and proactive maintenance,
limitations. AI algorithms can be computationally intensive, the proposed model endeavors to reduce downtime and
requiring substantial processing power and resources to analyze operational costs, optimizing the utilization of resources and
large datasets and make accurate predictions. Implementing such improving system efficiency.
techniques may pose computational challenges in certain power ➢ The model's adaptability to changing fault patterns and
system environments [2]. The effectiveness of AI models evolving power system conditions aims to ensure a robust and
heavily relies on the availability of vast and diverse fault datasets future-proof fault detection and protection system.
for training. In some cases, obtaining sufficient labeled data may ➢ The integration of AI-driven fault detection and protection
be challenging, hindering the model's performance. can serve as a stepping stone toward building intelligent and self-
healing power grids, capable of addressing faults autonomously
Certain AI models, like Neural Networks, operate as "black and efficiently.
boxes," making it difficult to comprehend how they arrive at
judgments. In safety-critical applications, this lack of Hence, integrating AI-driven fault detection and protection
interpretability could pose questions. Understanding the techniques into electric power systems holds great promise for
reasoning behind fault detection decisions is crucial for trust and improving reliability and operational efficiency. While the
confidence in the system [3]. Moreover, AI models can be proposed model offers numerous advantages, it is essential to
vulnerable to adversarial attacks, where malicious inputs address the associated challenges to ensure its successful
intentionally deceive the system and lead to incorrect fault implementation in real-world power system environments. With
detection or protection decisions. Ensuring the security and continued research and development, AI-driven fault detection
robustness of AI-driven techniques is a significant concern, and protection are poised to play a pivotal role in ensuring an
especially in critical infrastructure applications like electric enhanced and resilient electric power infrastructure.
power systems[4].
2. Related Works
To address these limitations, ongoing research is exploring
methods to enhance the interpretability and robustness of AI- In the field of fault detection and protection in electric
driven fault detection and protection systems. Techniques like power systems, numerous research works have been conducted,
explainable AI and adversarial training aim to make AI models utilizing various techniques such as rule-based methods, signal
more transparent and resilient against attacks. Hence, integrating processing, and traditional machine learning algorithms [6].
AI-driven fault detection and protection techniques holds great Early fault detection techniques relied on rule-based approaches,
promise for enhancing the reliability and stability of electric where predefined rules and heuristics were used to identify
power systems. These advanced algorithms offer accurate and specific fault conditions [7]. While these methods are simple and
real-time fault detection, proactive maintenance, and interpretable, they often struggle to handle complex fault
adaptability to changing conditions. However, challenges such patterns and adapt to varying system conditions [8]. Signal
as computational complexity, data availability, interpretability, processing methods, such as Fourier transform [9] and wavelet
and security must be carefully addressed to fully harness the analysis [10], have been employed to extract fault signatures
potential of AI in power system protection. Future developments from power system signals. Although these techniques offer
in this field will play a crucial role in building smarter and more valuable insights, they may encounter challenges with noise and
resilient power infrastructures, safeguarding against faults and variations in fault patterns. For defect detection and
ensuring continuous and reliable electricity supply to society [5]. classification, traditional ML methods such as DT [11] and
Support Vector Machines have been used [12]. While showing
The suggested model's main goal is to provide an extensive and promising results in some scenarios, these approaches may
reliable fault detection and protection system for electrical power struggle with highly nonlinear and complex fault patterns [13].
systems. The proposed model seeks to improve fault detection In recent years, AI-driven techniques [14], particularly Deep
accuracy by using AI algorithms capable of effectively Learning with Neural Networks [15], have gained popularity for
identifying subtle fault patterns that may be challenging for fault detection tasks. Neural Networks [16] can automatically
traditional methods. With the integration of AI, the proposed learn complex fault representations [17], leading to improved

294
 INTERNATIONAL JOURNAL of RENEWABLE ENERGY RESEARCH
R. Venkatasubramanian et al., Vol.14, No.2, June, 2024
accuracy and performance compared to traditional approaches 3. Base Models
[18].
We will be discussing the fundamental models used in the
Despite the progress made in fault detection and study work for fault detection and prevention in the electric
protection, the existing works have certain limitations. The power system: Logistic Regression, Decision Tree, SVM, and
interpretability of fault detection systems [19] is crucial in Random Forest.
safety-critical applications, making the lack of interpretability in
some existing fault detection methods an important concern. 3.1 Logistic Regression

Additionally, it's important to ensure real-time monitoring For binary classification tasks, the statistical technique of
and problem detection in intricate power grid networks [20], logistic regression is frequently utilized. It is used to forecast the
which calls for approaches that can scale to huge power systems existence or absence of a particular problem type based on input
with lots of parts and vast amounts of data streams [21]. Many parameters such line currents and voltages in the context of fault
machine learning algorithms require large and diverse fault detection and protection. The technique uses a logistic function
datasets for training, which may not always be readily available to represent the link between the input features and the likelihood
in real-world power system applications [22]. Moreover, of a defect occurring. It fits a linear decision boundary to the
traditional methods may struggle to adapt to changing system training data and maps the predicted probability to binary classes
conditions and new fault scenarios, potentially hindering their (e.g., No Fault or Fault). To reduce the discrepancy between
effectiveness [23]. anticipated and real fault labels during training, the model's
weights and biases are improved using a technique called
To address these limitations, the proposed AI-driven fault gradient descent. Logistic Regression is computationally
detection and protection model offers several advantages over efficient and interpretable, making it a good choice for simple
existing works. By leveraging AI techniques like Deep Forest and linearly separable fault detection tasks.
and Neural Networks, the proposed model can achieve higher
fault detection accuracy compared to rule-based and traditional Given a dataset with m samples and n features, the input
machine learning methods [24). The ability to learn complex features can be represented as a matrix X ∈ ℝm x n , where each
fault patterns enables more precise fault identification. Real-time row corresponds to a sample and each column corresponds to a
fault detection is a key advantage of the proposed model, feature. Let the binary target variable (fault or no fault) be
enabling prompt mitigation actions to prevent cascading failures represented as a vector y ∈ {0, 1}m . The logistic regression
and minimize system downtime, especially in safety-critical model estimates the probability of a fault occurrence
power system applications [25]. Unlike rule-based methods, the (P(y = 1|x)) using the sigmoid function given in Eq. 1.
proposed model possesses adaptive learning capabilities. It can
continuously learn from new fault data, adapt to changing power 1
(𝑃(𝑦 = 1|𝑥)) = (1 + 𝑒𝑥𝑝(−𝑧))
(1)
system conditions, and improve its performance over time [26].

The integration of AI-driven techniques [27] ensures the where z is the linear combination of the model parameters
model's robustness to handle variations and emerging fault (weights and biases) provided in Equation 2 and the input
patterns, making it future-proof and well-suited for dynamic features.
power system operations. The suggested model [28] can also 𝑧 = 𝛽0 + 𝛽1 ∗ 𝐼𝑎 + 𝛽3 ∗ 𝐼𝑐 + 𝛽4 ∗ 𝑉𝑎 + 𝛽5 ∗ 𝑉𝑏 + 𝛽6 ∗ 𝑉𝑐 (2)
contain interpretability approaches to offer insights into the
decision-making process for defect identification, offering In Eq. 3, the vectorized form is given:
clarity and explanations where necessary. The suggested model 𝑧= 𝑋∗ β (3)
[29] may be used to large-scale power systems with numerous
components using scaled AI approaches, enabling thorough and Where, β = [β₀, β₁, β₂, β₃, β₄, β₅, β₆] is the parameter vector.
effective problem identification [30]. To train the logistic regression model, the MLE method is used
to find the optimal values for β that maximize the likelihood of
Hence, the proposed AI-driven fault detection and the data given the model. In order to do this, the adverse log-
protection model offers distinct advantages over existing works, likelihood function provided by Eq. 4 must be minimized.
including improved accuracy, real-time monitoring, 𝐽(𝛽) = −1/𝑚 ∗ ∑[𝑦 ∗ 𝑙𝑜𝑔(𝑃(𝑦 = 1|𝑥)) + (1 − 𝑦) ∗
adaptability, interpretability, and scalability. The suggested 𝑙𝑜𝑔(1 − 𝑃(𝑦 = 1|𝑥)) (4)
model has the potential to considerably improve the stability and
dependability of electric power systems by resolving the where m is the dataset's sample count. The model parameters β
shortcomings of conventional approaches, paving the way for are updated iteratively using optimization algorithms like
smarter and more robust power grid networks [31]. The gradient descent until convergence.
integration of Deep Forest and Neural Networks, along with
adaptive learning capabilities, ensures the model's ability to 3.2 Support Vector Machines (SVM)
handle complex fault scenarios and adapt to changing
conditions, making it a promising method for identifying and SVM are powerful supervised learning techniques used to
preventing faults in electric power networks [32]. Future solve regression and classification issues. Finding the proper
developments in this field may further enhance the model's hyperplane in a high-dimensional feature space to divide the data
performance, ultimately contributing to more robust and secure points of different fault types is the aim of support vector
power system operations [33]. machines (SVM). SVM seeks to categorize various fault types

295
 INTERNATIONAL JOURNAL of RENEWABLE ENERGY RESEARCH
R. Venkatasubramanian et al., Vol.14, No.2, June, 2024
in the context of fault detection by identifying the optimum Decision Trees may represent non-linear connections
decision boundary that optimizes the difference in class margins. between characteristics and fault kinds and are comprehensible.
SVM is capable of handling non-linearly separable data by When the tree gets too deep, they are vulnerable to overfitting.
employing kernel functions to translate the data to a space with To reduce overfitting and boost generalization, pruning
more dimensions. approaches and ensemble methods like Random Forest are
utilized.
The kernel trick allows SVM to efficiently solve complex
fault classification problems and make it robust against noise. 3.4 Random Forest
SVM is effective in handling large datasets with high
dimensionality and can generalize well with proper tuning of Multiple Decision Trees are used in Random Forest, an
hyperparameters. The SVM model builds a hyperplane that ensemble learning technique, to increase the fault detection
optimizes the space between the samples from distinct classes model's overall performance and resilience. In Random Forest, a
that are closest to the decision boundary. Eq. 5 contains the SVM number of decision trees are constructed using various random
model's decision function. subsets of the training data and characteristics. The projected
fault type is then put to a vote by each tree, and a decision is
reached based on a majority vote. Random Forest lowers the
f(x) = sign(∑[αi ∗ yi ∗ K(x, xi )] + b) 5) chance of overfitting and improves the model's accuracy and
generalizability by averaging the predictions of numerous trees.
where 𝛼𝑖 and 𝑦𝑖 are the Lagrange multipliers and fault type labels It can handle correlations between input characteristics and fault
for the support vectors 𝑥𝑖 , respectively. 𝐾(𝑥, 𝑥𝑖 ) is the kernel kinds that are both linear and non-linear. For problem detection
function that computes the similarity between input data x and in electrical power systems, Random Forest is a common choice
support vectors 𝑥𝑖 . In Eq. 6, the decision function becomes: since it is less susceptible to hyperparameter adjustment and
noise. Multiple Decision Trees are trained on bootstrap samples
f(x) = sign(αᵀ ∗ K(x, X) + b) (6) of the data (sampling with replacement) to create a Random
Forest. Additionally, each decision node only takes into account
Where, α = [α1 , α2 , . . . , αₘ] is the vector of Lagrange a random subset of characteristics for each split. This
multipliers, and K(x, X) is the kernel matrix, where each element randomization boosts the ensemble's variety and decreases the
𝐾(𝑥, 𝑥𝑖 ) represents the similarity between input data x and connection between trees.
support vector 𝑥𝑖 . To train the SVM model, the quadratic
programming problem is solved to determine the ideal values of The following is the decision-making procedure for a fresh
α in order to optimize the margin and meet the requirements sample x:
∑(αᵢ ∗ yᵢ) = 0 and 0 ≤ αᵢ ≤ C, where C is the 1. Go through every tree in the Random Forest with x.
regularization parameter. 2. Using a majority vote (for classification) or an average (for
regression), combine all of the trees' predictions.
3.3 Decision Tree 3. The average (regression) or majority vote (classification) of
all tree forecasts is the final prediction.
A non-linear, interpretable technique called a decision tree Random Forests are robust against overfitting and noise and can
is employed for both classification and regression problems. In handle large datasets with high dimensionality. They often yield
order to generate a tree-like structure of decision nodes and leaf higher accuracy compared to individual Decision Trees, making
nodes for fault detection, a Decision Tree splits the data them a popular choice for fault detection and protection tasks.
recursively depending on the feature values. Each leaf node In the research paper, we have integrated these base
indicates the type of projected defect, and each decision node models (Logistic Regression, Decision Tree, Random Forest,
represents a test on one of the input characteristics. Climbing up and SVM) to perform fault detection and protection in the
the hierarchy from the root nodes to a leaf node in the decision- electric power system. By leveraging the strengths of each
making process is dependent on the values of the input attributes. model, the hybrid approach demonstrates improved accuracy
The feature that optimizes the information gain or Gini impurity and reliability in identifying and classifying different fault types.
is used to iteratively divide the data into subgroups and build a The experimental results reveal that the hybrid model
decision tree. Throughout the process, a stopping condition like outperforms individual base models and existing techniques,
a maximum depth or a minimum number of samples per leaf may showcasing its potential for enhancing the reliability and
be achieved at any point. stability of electric power systems [34-38].

The decision-making process for a new sample x is as

follows:
1. Start from the root node.
2. Test the value of a specific feature at the root node.
3. If the feature value meets the test criteria, go to the left child; 4. Integrated AI-driven Fault Detection and Protection
otherwise, move to the right child. Model
4. Continue performing steps 2 and 3 until you reach a leaf node.
To improve problem detection and prevention in electric
5. The decision tree's output is the projected failure category at
power components and systems, a hybrid model is being
the leaf node.
suggested that integrates AI-driven methodologies [39]. Figure

296
 INTERNATIONAL JOURNAL of RENEWABLE ENERGY RESEARCH
R. Venkatasubramanian et al., Vol.14, No.2, June, 2024
1 depicts the proposed system's general operation. The model is 4.1 Deep Forest Feature Extractor
composed of the following components:

Deep Forest is employed as the feature extractor to capture Where, 𝑦𝑖 is the fault label of the i-th sample, and 𝑍𝑖 is its
informative representations from the input fault data. It utilizes corresponding feature vector in Z. The optimal w and b are
a cascade of Random Forest classifiers to learn essential fault obtained by solving the following optimization problem in Eq.
characteristics. Assume that X is the input fault data matrix with (9):
dimensions (m, n), where m is the sample count and n is the
1
feature count. The Deep Forest feature extractor processes X 𝑚𝑖𝑛𝑖𝑚𝑖𝑧𝑒 (2) = ‖𝑤‖2 (9)
through a series of Random Forest classifiers to obtain the
transformed feature matrix Z with dimensions (m, k), where k is 𝑠𝑢𝑏𝑗𝑒𝑐𝑡 𝑡𝑜: 𝑦𝑖 = (𝑤 𝑇 ∗ 𝑍𝑖 + 𝑏) ≥ 1 𝑓𝑜𝑟 𝑎𝑙𝑙 𝑖 = 1,2. ., (10)
the number of extracted features in Eq. (7).
4.3 Neural Networks (NN) Classifier
Z = DeepForestFeatureExtractor(X) (7)
The feature matrix Z is also used as input to a Neural
4.2 Support Vector Machines (SVM) Classifier Network classifier. The NN model comprises multiple
interconnected layers with non-linear activation functions,
The extracted features Z from the Deep Forest component
enabling it to learn complex fault patterns from the extracted
are then fed into an SVM classifier. SVM aims to find the
features. Let f(x, W) be the NN model with parameters W that
optimal hyperplane that effectively separates feature vectors
maps the input feature vector Z to the output fault class
corresponding to different fault classes. Given the feature matrix
probabilities. The classification process involves forward
Z with dimensions (m, k) and the corresponding fault labels y
propagation to calculate the output probabilities and
with dimensions (m, 1), the SVM classifier finds the hyperplane
backpropagation to update the weights to minimize the
w and bias b that maximize the margin between different fault
classification loss in Eq. (11).
classes while satisfying the constraint in Eq. (8):
y_pred = f(Z, W) (11)
𝑦𝑖 = (𝑤 𝑇 ∗ 𝑍𝑖 + 𝑏) ≥ 1 𝑓𝑜𝑟 𝑎𝑙𝑙 𝑖 = 1,2. . , 𝑚 (8)

Data set
Preparation Pre-
12000 labeled data processing
points of power
system scenario

Training Validation

Performance
Evaluation
Base Models Is model
is
trained
Integrated AI Testing
driven dataset

Analysis of Classification Process

Data Training
Dataset

Finding
out faults
in power
system

Classify Classify faulty

healthy system
system

Determine best ML
method for fault

Fig.1. Overall flow diagram for proposed system architecture

297
 INTERNATIONAL JOURNAL of RENEWABLE ENERGY RESEARCH
R. Venkatasubramanian et al., Vol.14, No.2, June, 2024
5. Dataset Description and Experimentation

The information made available is the result of a power - G: Generator status (binary value: 0 for off, 1 for on).
system simulation run in MATLAB to examine and research
failure scenarios. Four generators make up the power system - C, B, A: Fault indicators for different fault types (binary values:
arrangement, and each one produces energy at a voltage of 0 for no fault, 1 for fault occurrence).
11 × 103 V. These generators are placed at the transmission
line's ends. Transformers are used between the generators to - Ia, Ib, Ic: Current values in lines A, B, and C, respectively.
mimic different fault situations that might arise at the
transmission line's middle. - Va, Vb, Vc: Voltage values in lines A, B, and C, respectively.

There are two primary scenarios in the simulation: typical - Fault_Type: A categorical label indicating the specific type of
operating settings and different fault states. Line Voltages and fault that occurred during the simulation.
Line Currents at the output side of the power system are
We conducted a comprehensive analysis of various faults
monitored and recorded under these circumstances [40]. The
that can occur in an electric power system. These faults are
dataset contains approximately 12,000 data points, and each data
categorized into different types to facilitate a detailed
point is labeled with a specific "Fault_Type," indicating the type
investigation and understanding of their impact on system
of fault that was simulated during that particular instance.
performance. The fault categories considered are as follows:
The dataset is organized with the following columns:

5.1 No Fault (Healthy System) as shown in Figure 2 and 3. The current and voltage waveforms
are 120 degrees out of phase, with the maximum current ranging
In the absence of any faults, the power system operates from approximately +100 to -100 Amperes and the voltage
under normal conditions. During this scenario, both the Voltage fluctuating between +0.5 pu to -0.5 pu.
and Current graphs exhibit symmetrical and sinusoidal behavior

Fig.4. Current Behavior in System that is flawed with a Line A

Fig.2. Current Behavior in No Fault (Healthy System)
to Ground Fault

Fig.3. Voltage Behavior in No Fault (Healthy System) Fig.5. Voltage Behavior with Line A to Ground Fault in Faulty
System
5.2 Faulty System with Line A to Ground Fault
5.3 System Fault with Line A, Line B, and Ground Fault
When a fault occurs between Line A and the ground, the
system experiences an asymmetrical fault condition as shown in From Figures 6 and 7, the fault occurs between Line A,
Figure 4 and 5. The current in Line A significantly increases, Line B, and the ground. The fault currents in both Line A and
surging to approximately 10 times its normal value, reaching Line B experience substantial increases, deviating from their
around 1000 Amperes. Simultaneously, the voltage in the system normal values. The voltage levels in the system also undergo
reduces due to the fault occurrence. reductions.

298
 INTERNATIONAL JOURNAL of RENEWABLE ENERGY RESEARCH
R. Venkatasubramanian et al., Vol.14, No.2, June, 2024

Fig.10. Current System Behavior with Line A, Line B, and Line

Fig.6. Current System Behavior with Line A, Line B, and C Faults
Ground Fault

Fig.11. Voltage Behavior in the Line A, Line B, and Line C Fault

Fig.7. Voltage Behavior with Line A, Line B, and Ground Fault
System
in Faulty System
5.6 System Fault involving Line A, Line B, Line C, and Ground
5.4 Line B to Line C Fault in Faulty System
From Figures 12 and 13, faults occur in all three lines and
In Figures 8 and 9, a fault between Line B and Line C
are connected to the ground. This results in an unbalanced and
results in an asymmetrical fault condition. The current in Line B
severe fault condition. The fault currents in all three lines and the
and Line C may experience significant deviations from their
ground experience substantial deviations from their normal
normal values. Additionally, the voltage levels in the system are
values. The voltage levels in the system are also significantly
affected due to the fault.
impacted.

Fig.8. Current System Behavior with Line B to Line C Fault Fig.12. Current System Behavior with Line A, Line B, Line C,
and Ground Faults

Fig.9. Voltage Behavior with Line B to Line C Fault in Faulty Fig.13. Voltage Behavior in a Defective System with Lines A,
System B, C, and Ground
5.5 System Fault involving Line A, Line B, and Line C Faults Through an in-depth analysis of these different fault
categories, we aim to provide valuable insights into fault
When faults occur in all three lines, the system encounters detection and protection techniques in electric power systems
a complex fault scenario. The fault currents in all three lines [41-. Our suggested model delivers improved fault detection
exhibit abnormal behavior, and the voltage levels are accuracy and real-time monitoring by utilizing AI-driven
significantly affected by the faults are shown in Figures 10 and techniques as Deep Forest, Support Vector Machines, and
11. Neural Networks. The integration of interpretability techniques
ensures transparency in decision-making, making our model a
reliable and robust solution for enhancing electric power system
reliability and safety [45]. By addressing the limitations of

299
 INTERNATIONAL JOURNAL of RENEWABLE ENERGY RESEARCH
R. Venkatasubramanian et al., Vol.14, No.2, June, 2024
existing works, our proposed model can effectively detect faults, system reliability and safety. By integrating multiple AI-driven
mitigate potential risks, and maintain the stability of power techniques, our proposed model overcomes the limitations of
components and systems, ultimately leading to improved electric individual models and offers a robust and reliable solution for
power system reliability [46]. fault detection in electric power components and systems.

6. Results and Discussion 7. Conclusion and Future Works

Our dataset was used to assess the findings and arguments We provided a thorough analysis of fault detection and
on the effectiveness of various defect detection methods. The protection methods for improving the dependability of the
models that are evaluated are Logistic Regression, Decision electric power system in this research article. To effectively
Tree, SVM, Random Forest, and our proposed Hybrid Model identify and categorize distinct defect kinds, A range of AI-
that integrates AI-driven fault discovery techniques. driven models, including Random Forest, SVM, Decision Trees,
Logistic Regression, and our own proposed Hybrid Model, were
Table1. Scores obtained for each model's training accuracy and constructed and evaluated.
model accuracy are summarized.
The evaluation's findings proved that our suggested hybrid
S.No Model Training Testing model was better than the conventional models. The Hybrid
Accuracy Accuracy Model achieved an outstanding training accuracy of 99.12% and
1 Logistic Regression 76.52% 74.19% a remarkable model accuracy score of 98.57%. This exceptional
2 Support Vector Machine 90.19% 89.57% performance can be attributed to the model's innovative
3 Decision Tree 93.23% 90.55% integration of Deep Forest, Support Vector Machines, and
4 Random Forest 95.47% 93.32% Neural Networks. By leveraging Deep Forest as a feature
5 Hybrid Model 99.12% 98.57% extractor and combining SVM and NN for classification and
decision-making, our proposed model excelled in accurately
detecting various fault categories, providing real-time
From Table 1, we observe that the Hybrid Model achieved the monitoring, and improving electric power system reliability. The
highest training accuracy of 99.12% and an impressive model advantages of our Hybrid Model lie in its ability to handle
accuracy score of 98.57%. The model's ability to combine the complex fault patterns and achieve high accuracy in fault
benefits of several AI-driven methodologies, such as Deep detection. Our model addresses the shortcomings of individual
Forest, Support Vector Machines, and Neural Networks, is what models and provides a more substantial and dependable solution
accounts for its extraordinary performance. Despite being a for fault detection and protection in electric power components
simple model, logistic regression showed acceptable and systems by combining a number of AI-driven
performance, with a training accuracy score of 76.51% and a methodologies. The model's interpretability promotes decision-
model accuracy score of 74.19%. However, it shows limitations making transparency, facilitating operator and engineer
in handling complex fault patterns due to its linear nature. understanding of and trust in the defect detection process.
Support Vector Machines (SVM) fared substantially better,
scoring 89.57% for model accuracy and 90.19% for training For future development, we aim to further enhance the
accuracy. SVM was a good option for defect detection jobs Hybrid Model by exploring additional AI-driven techniques,
because of its capacity to handle non-linear correlations between such as Deep Learning and Reinforcement Learning. The
features. With a training accuracy of 93.23% and a model incorporation of these advanced techniques may improve fault
accuracy score of 90.55%, Decision Tree showed good results. detection accuracy even further and enable self-learning
Its ability to create interpretable decision rules made it useful for capabilities for adaptive fault protection. We also intend to
understanding the fault detection process. An ensemble method gather larger and more varied datasets in order to verify the
called Random Forest showed increased accuracy, with a model's functionality across a range of failure scenarios and
training accuracy score of 95.47% and a model accuracy score systems setups. The scalability and adaptability of the model will
of 93.32%. By aggregating the outputs of multiple decision trees, be essential for its practical implementation in large-scale power
Random Forest enhanced fault detection capabilities. systems.

However, the most noteworthy performance was achieved

by our proposed Hybrid Model. The Hybrid Model's Declaration:
combination of Deep Forest, Support Vector Machines, and
Neural Networks resulted in exceptional fault detection Ethics Approval and Consent to Participate:
accuracy. By leveraging Deep Forest as a feature extractor and No participation of humans takes place in this
utilizing SVM and NN for classification and decision-making, implementation process
the model showcased remarkable performance and outperformed
all other models in both training accuracy and model accuracy Human and Animal Rights:
score. Overall, the Hybrid Model's outstanding performance No violation of Human and Animal Rights is involved.
makes it a promising solution for enhancing fault detection and
protection in electric power systems. Its ability to accurately Funding:
classify various fault types and provide real-time monitoring No funding is involved in this work.
capabilities showcases its potential for ensuring electric power
Data availability statement:

300
 INTERNATIONAL JOURNAL of RENEWABLE ENERGY RESEARCH
R. Venkatasubramanian et al., Vol.14, No.2, June, 2024
Data sharing not applicable to this article as no datasets Acknowledgement:
were generated or analyzed during the current study
There is no acknowledgement involved in this work.
Conflict of Interest:
Conflict of Interest is not applicable in this work.
Authorship contributions:
All authors are contributed equally to this work

References International Journal of Electrical Power & Energy

Systems, vol. 135, pp. 106611, 2021.
[1] X. Z. Liu, Y. W. Zhang, and Q. Z. Chen, "An AI-driven
approach for fault detection in power systems using deep [10] F. L. Chen, Y. J. Zhang, and Y. H. Wang, "A
learning," Electric Power Systems Research, vol. 121, comparative study of fault detection techniques for
pp. 102-112, 2020. electric power systems using artificial neural networks,"
Electric Power Systems Research, vol. 198, pp. 107279,
[2] A. B. Smith, M. J. Johnson, and C. R. Lee, "Enhancing 2021.
power system reliability through fault protection
optimization using genetic algorithms," IEEE [11] S. B. Rai, R. S. Meena, and A. K. Singh, "Advanced
Transactions on Power Systems, vol. 36, no. 3, pp. 1847- fault detection and protection in smart grids using hybrid
1856, 2021. fuzzy logic and genetic algorithms," Sustainable Energy
Technologies and Assessments, vol. 52, pp. 101432,
[3] S. K. Gupta, P. N. Sharma, and R. S. Kapoor, "Fault 2022.
detection and protection in microgrids using smart grid
technologies," Renewable Energy, vol. 160, pp. 1001- [12] H. J. Wang, Q. Y. Zhu, and W. X. Li, "Machine
1010, 2021. learning-based fault detection and protection for
microgrids with renewable energy integration," IEEE
[4] M. A. Rahman, F. H. Khan, and S. M. Ali, "A novel Transactions on Smart Grid, vol. 13, no. 2, pp. 1058-
approach for fault type classification in power systems 1067, 2022.
using wavelet transform and machine learning,"
International Journal of Electrical Power & Energy [13] R. S. Sharma, S. D. Gupta, and A. K. Jain, "An AI-
Systems, vol. 132, pp. 107173, 2021. driven approach for fault classification in power
components using wavelet transform and extreme
[5] N. C. Patel, S. A. Desai, and K. V. Shah, "Intelligent fault learning machines," Electric Power Systems Research,
detection and protection in power distribution systems vol. 238, pp. 107113, 2022.
using fuzzy logic and neural networks," Electric Power
Components and Systems, vol. 49, no. 6, pp. 612-623, [14] G. H. Kim, K. J. Lee, and S. J. Park, "A novel hybrid
2021. deep learning-based fault detection and classification
technique for electric power systems," Electric Power
[6] L. H. Wu, Y. L. Wang, and X. Q. Li, "A hybrid AI Systems Research, vol. 260, pp. 107115, 2022.
approach for fault detection in smart grids using support
vector machines and particle swarm optimization," [15] S. H. Liu, Y. K. Chen, and H. H. Wu, "Fault detection
Sustainable Energy Technologies and Assessments, vol. and protection in distribution networks using machine
46, pp. 101508, 2021. learning and internet of things," Electric Power
Components and Systems, vol. 51, no. 11, pp. 1127-
[7] K. M. Chen, S. L. Chen, and C. H. Wang, "Fault 1137, 2022.
classification in electric power components using
ensemble learning techniques," International Journal of [16] M. R. Gupta, P. N. Patel, and R. K. Sharma, "A
Electrical Engineering Education, vol. 58, no. 4, pp. 332- comparative study of fault detection algorithms for
345, 2021. electric power components using ensemble learning
techniques," International Journal of Electrical
[8] Q. T. Nguyen, V. L. Tran, and T. T. Le, "Enhancing Engineering Education, vol. 59, no. 5, pp. 391-404, 2022.
power system reliability through fault detection and
protection using a hybrid artificial immune algorithm," [17] A. M. Sharma, S. P. Singh, and V. K. Verma, "A
Electric Power Systems Research, vol. 192, pp. 106887, hybrid intelligent approach for fault detection in power
2021. systems using adaptive neuro-fuzzy inference systems,"
Sustainable Energy Technologies and Assessments, vol.
[9] M. J. Lee, S. Y. Kim, and H. G. Park, "A machine 64, pp. 101622, 2023.
learning-based fault detection and classification
approach for power systems using clustering algorithms," [18] J. D. Kim, S. K. Park, and W. J. Kang, "An enhanced
fault detection and classification approach for power

301
INTERNATIONAL JOURNAL of RENEWABLE ENERGY RESEARCH
R. Venkatasubramanian et al., Vol.14, No.2, June, 2024
systems using machine learning ensemble techniques," Measurement Units," Electric Power Systems Research,
International Journal of Electrical Power & Energy vol. 190, 106515, 2021.
Systems, vol. 169, pp. 107582, 2023.
[29] X. Zhang, C. Deng, Z. Wang, and H. Wang, "Fault
[19] H. R. Lee, K. H. Yoo, and C. S. Lim, "A hybrid fault Detection and Identification in MMC-HVDC Systems
detection and protection technique for microgrids using using Convolutional Neural Networks," IEEE
deep reinforcement learning," Electric Power Systems Transactions on Power Delivery, vol. 36, no. 4, pp. 2043-
Research, vol. 294, pp. 107439, 2023. 2052, 2021.

[20] P. G. Gupta, S. S. Agarwal, and R. K. Jain, "AI-driven [30] D. N. V. S. L. S. Indira, Rajendra Kumar Ganiya, P.
fault detection and protection for power systems with Ashok Babu, A. Jasmine Xavier, L. Kavisankar, S.
renewable energy integration using neural networks and Hemalatha, V. Senthilkumar, T. Kavitha, A. Rajaram,
genetic algorithms," Renewable Energy, vol. 218, pp.
Karthik Annam, and Alazar Yeshitla, “Improved
1502-1512, 2023.
Artificial Neural Network with State Order Dataset
[21] S. A. Gupta, A. B. Sharma, and R. K. Singh, "An IoT- Estimation for Brain Cancer Cell Diagnosis”, BioMed
based approach for fault detection and protection in smart Research International, vol. 2022, 10 pages, 2022.
grids using machine learning algorithms," Electric Power [31] K. Kalaivani, PR. Kshirsagarr, J. Sirisha Devi, SR.
Systems Research, vol. 308, pp. 107436, 2023. Bandela, I. Colak, J.Nageswara Rao, A.Rajaram,
“Prediction of biomedical signals using deep learning
[22] H. J. Patel, S. R. Kumar, and V. M. Sharma, "A novel
techniques,”Journal of Intelligent & Fuzzy Systems,1-
fault classification technique for electric power
components using wavelet transform and support vector 4,2023.
machines," International Journal of Electrical
Engineering Education, vol. 62, no. 4, pp. 350-362, 2023. [32] J. Gao, Y. Xu, C. Zhang, and W. Wang, "A Fast and
Accurate Fault Detection and Localization Method in
[23] L. M. Agarwal, R. S. Verma, and S. K. Gupta, "Fault Microgrids using Synchrophasor Data," Electric Power
detection and protection in smart grids using ensemble Components and Systems, vol. 49, no. 3, pp. 285-295,
learning and data fusion techniques," Sustainable Energy 2021.
Technologies and Assessments, vol. 80, pp. 101969,
2023. [33] R. Kalpana, V S, R. Lokanadham, K. Amudha, GN,
Beena Bethel,AK Shukla, PR, Kshirsagar, A. Rajaram,
[24] C. S. Kumar, A. K. Gupta, and M. S. Jain, "Enhancing “Internet of Things (IOT) Based Machine Learning
power system reliability through fault detection and
Techniques for Wind Energy Harvesting,” Electric
protection using hybrid genetic algorithms and particle
swarm optimization," Electric Power Components and Power Components and Systems, 14:1-7, December
Systems, vol. 54, no. 8, pp. 743-754, 2023. 2023.

[34] V. Ş. Ediger, G. Kirkil, E. Çelebi, M. Ş. Ucal, and Ç.

[25] S. P. Singh, A. K. Sharma, and V. N. Verma, "A K. Çin, “Turkish Public Preferences for Energy”, Kadir
comparative study of fault detection techniques for Has University, Center for Energy and Sustainable
electric power components using deep learning and fuzzy Development, 2021.
logic," International Journal of Electrical Power &
Energy Systems, vol. 190, pp. 107175, 2023. [35] C. Yan, M. Qu, Y. Chen, and M. Feng, "Snow
removal method for self-heating of photovoltaic panels
[26] D. Wang, M. Zhang, Y. Zheng, H. Wang, and Z. Li, and its feasibility study", Solar Energy, vol. 206, pp. 374-
"An Adaptive Multi-Rate Protection Scheme for 380, 2020.
Distribution Systems Based on Fault Classification and
Faulted Phase Selection," IEEE Transactions on Power
[36] Kilic, and Ahmet, "Plug and Charge solutions with
Delivery, vol. 35, no. 5, pp. 2505-2514, 2020.
vehicle-to-grid communication," Electric Power
Components and Systems, (2023): 1-29.
[27] X. Liu, Q. Chen, C. Yuen, and P. Chau, "Transient-
based Fault Detection and Localization in HVDC Grids
with Consideration of Converter Delays," IEEE [37] P Maske, P A Chel, P K Goyal, and D G Kaushik,
Transactions on Power Systems, vol. 35, no. 2, pp. 1022- “Sustainable Perspective of Electric Vehicles and Its
1033, 2020. Future Prospect. Jsmpm”, Dec. 2021; vol. 1, no. 1, pp.
17–32.
[28] Y. Xiang, W. Shao, H. Zhang, and M. A. S. Masoum,
"Adaptive Fault Location in Distribution Systems with [38] P. Ashok Babu, JL. Mazher Iqbal,S.Siva Priyanka, M.
Renewable Energy Integration Using Phasor Jithender Reddy, G. Sunil Kumar, R.Ayyasamy, “Power
control and optimization for power loss reduction using

302
INTERNATIONAL JOURNAL of RENEWABLE ENERGY RESEARCH
R. Venkatasubramanian et al., Vol.14, No.2, June, 2024
deep learning in microgrid systems,” Electric Power International Conference on Civil Aviation Safety and
Components and Systems.;52(2):219-32, January 2024. Information Technology (ICCASIT) (pp. 894-897),
IEEE, October 2021.
[39] B. Sarsembayev, D. Zholtayev, and T. D. Do,
"Maximum Power Tracking of Variable-Speed Wind [44] A. Sahbani, K. Cherif ,KB. Saad, “Multiphase
Energy Conversion Systems based on a Near-Optimal Interleaved Bidirectional DC-DC Converter for Electric
Servomechanism Control System", Optim. Control Appl. Vehicles and Smart Grid Applications,” International
Methods, vol. 43, no. 3, pp. 904-924, 2022. Journal of Smart Grid-ijSmartGrid, 28;4(2):80-7, June
2020.
[40] TG.Amaral, VF. Pires, D. Foito, AJ. Pires , JF.
Martins, “Fault Detection and Diagnosis Technique for a [45] H. Shekhar , C. Bhushan Mahato , SK. Suman ,S.
SRM Drive Based on a Multilevel Converter Using a Singh , L. Bhagyalakshmi ,M. Prasad Sharma , B. Laxmi
Machine Learning Approach,” In2023 12th International Kantha, SK. Agraharam , A. Rajaram, “Demand side
Conference on Renewable Energy Research and control for energy saving in renewable energy resources
Applications (ICRERA,) (pp. 40-45), IEEE, August using deep learning optimization,” Electric Power
2023. Components and Systems. 26;51(19):2397-413,
November 2023.
[41] HS.Alnafee, “Robustness Analysis of ELM-based
Fault Detection in PV Systems,” International Journal of [46] H. Shekhar , C. Bhushan Mahato , SK. Suman ,S.
Smart Grid-ijSmartGrid. 22;7(4):189-99, December Singh , L. Bhagyalakshmi ,M. Prasad Sharma , B. Laxmi
2023. Kantha, SK. Agraharam , A. Rajaram, “Demand side
control for energy saving in renewable energy resources
[42] A. AlKassem , M. Al Ahmadi, A. Draou, “Modeling using deep learning optimization,” Electric Power
and simulation analysis of a hybrid PV-wind renewable Components and Systems. 26;51(19):2397-413,
energy sources for a micro-grid application” In2021 9th November 2023.
International Conference on Smart Grid (icSmartGrid,)
(pp. 103-106), IEEE, June 2019.

[43] H. Xiaotao , Y. Qiang , B. Ou , W. Shangjie, Z.

Weijia, Y. Shuai, L. Yajin, “Operation and Maintenance
System of Electric Vehicles’ Charging and Discharging
Facilities Based on Repository” In2021 IEEE 3rd .

303