0% found this document useful (0 votes)
14 views

DC microgrid protection

DC microgrid protection

Uploaded by

HR Habib
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
14 views

DC microgrid protection

DC microgrid protection

Uploaded by

HR Habib
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 17

Computers and Electrical Engineering 120 (2024) 109683

Contents lists available at ScienceDirect

Computers and Electrical Engineering


journal homepage: www.elsevier.com/locate/compeleceng

Enhancing robustness of DC microgrid protection against weather


intermittency and DER outage using optimal adaboost technique
Raushan Kumar a,b,* , Amit Kumar a, S.K. Parida b
a
Department of Electrical and Electronics Engineering, Sershah Engineering College Sasaram, Sasaram 821113, India
b
Department of Electrical Engineering, Indian Institute of Technology Patna 801106, India

A R T I C L E I N F O A B S T R A C T

Keywords: The swift integration of distributed generations (DG) in microgrids has spurred the advancement
DC microgrid of microgrid technology, promising efficient utilization of renewable energy sources. However,
Protection technique ensuring microgrid protection encounters significant challenges due to variable operational
Distributed energy resources
modes and the limited inertia of DG units. Existing protection solutions based on pre-specified
Adaboost classifier
Stationary wavelet transform
threshold settings quite often maloperate under the dynamic operating conditions caused due
Photovoltaic intermittency to network reconfiguration and intermittency of renewable Distributed Energy Resources (DERs).
To address such challenges, a novel protection scheme is proposed, which exploits the inherent
characteristics of Stationary Wavelet Transform (SWT) with Optimal Adaboost Ensemble (OAE)-
based technique. The proposed hybrid approach involves extraction of discriminatory attributes
characterizing fault behaviour through SWT and generalized input-output mapping capability of
OAE, thereby forming a robust foundation for fault analysis. The hybrid protection method ex­
hibits resilience in detecting and categorizing DC microgrid faults, addressing DER outages,
weather uncertainties, diverse loading scenarios and other operating conditions. The proposed
approach has been found to be outperforming in terms of accurately identifying and classifying
fault types, by achieving an overall accuracy of 99.44 % as compared to the Bagging Tree and
Random subspace classifiers having overall accuracy below 90 %. Compared to other existing
methods in the literature, the proposed approach has been proved to be fault resilient against the
PV intermittency and DER outage.

1. Introduction

In recent times, the rapid proliferation of distributed generations has attracted the attention towards wider acceptance of microgrid
technology, offering an enticing pathway for effectively harnessing renewable energy sources [1]. Over the last few decades, AC and
DC microgrids have significantly attracted the attention of researchers due to rising concern towards minimizing the carbon emission
and advancement in the power electronic interface to harness the green energy resources. Unlike AC distribution networks, DC
microgrids possess various advantages, including enhanced efficiency, reliability, flexibility, and better integration of renewable
energy sources, making them an appealing option for diverse power distribution and energy management applications [2,3].
Despite the numerous benefits associated with DC microgrids, significant protection challenges hinder the realization of their
potential advantages [4,5]. Thus, with wider acceptance of DC microgrid, ensuring suitable protection is also important aspect to

* Corresponding author at: Department of Electrical and Electronics Engineering, Sershah Engineering College Sasaram, Sasaram 821113, India.
E-mail address: [email protected] (R. Kumar).

https://doi.org/10.1016/j.compeleceng.2024.109683
Received 9 May 2024; Received in revised form 23 August 2024; Accepted 8 September 2024
Available online 20 September 2024
0045-7906/© 2024 Elsevier Ltd. All rights are reserved, including those for text and data mining, AI training, and similar technologies.
R. Kumar et al. Computers and Electrical Engineering 120 (2024) 109683

provide seamless and secured operation. However, various challenges pertaining to the protection of DC microgrid including divergent
variation in the generation profile and limited inertia of DG units has created hindrance in achieving the dependability and swiftness of
microgrid protection. These challenges stem from the post-fault DC current profile, which substantially exceeds the nominal current
during fault events. Additionally, the absence of zero crossing points in DC systems further complicates the design of suitable pro­
tection techniques [6,7]. Moreover, DER outages in DC microgrids can lead to significant fluctuations in fault current levels and re­
versals in the direction of current flow, underscoring the need for robust and reliable protection schemes capable of promptly detecting
and responding to various fault conditions and operational challenges in real-time. The dynamic nature of renewable DERs, such as
PV-based generation, due to uncertainty in solar irradiance, may result in fluctuations in voltage-current profiles [8]. The stochastic
characteristics of PV profiles necessitate continuous adaptation of traditional overcurrent relay threshold settings, further compli­
cating the protection task.
In this regard, the development of an efficient protection strategy is sought, which can detect and classify the pole-pole and pole-
ground faults accurately and reliably, with immunity to the contingencies including DER outages and weather intermittency of PV-
based DERs.
The microgrid protection strategies documented in the literature are mostly based on fundamental components. For instance, in
[9], the variations in current frequency components has been considered as fault identification indicators, while the protection
approach proposed in [10] is based on directional overcurrent relays, leveraging the impedance of the negative sequence element for
fault detection. Moreover, a new differential protection scheme for inverter-interfaced microgrids (IIMGs) based on the signed cor­
relation technique is introduced in [11] whereas a voltage-based relay type for microgrid protection is explored in [12].
Despite considerable advancements, the reported methods are often unable to sense the low current level of power electronics and
lack the required relaying speed for effective fault isolation. The prompt action and isolation of faulty feeder is expected from the
protective algorithm to prevent damage to power electronics and networks thereby ensuring the resilience of microgrids. The transient
period of faults, lasting a few milliseconds [13], contains valuable fault information that can be used for fault identification. With
regard to performing fault detection in DC microgrids, the protection strategy can be generalized into two types: communication-based
and local. Communication-based methods involves data collection from both ends of a connection to identify faults, and relies on
communication infrastructures. Conversely, local techniques detect faults at breaker locations using local relays, offering resilience
against communication-related issues.
Conventional protection devices involving overcurrent relays often face the difficulty of threshold setting adaptively for dynamic
conditions. The conventional fault detection methods also face challenges in rapidly discharging fault currents on the DC side within
milliseconds, diminishing fault current magnitudes and making the detection methods ineffective. To comprehensively assess existing
DC microgrid protection methods, a wide volume of works has been documented in the literature. Techniques such as traveling wave-
based protection [14], fault detection through fault current derivative analysis [15], and adaptive empirical wavelet transform signal
decomposition [16] have been explored, each with its limitations. Furthermore, most of these techniques rely on post-fault current and
voltage magnitudes, potentially increasing costs and compromising protection reliability in the event of communication failure.
Additionally, these techniques often fail to address issues such as DER outages, intermittency in PV-based DERs, internal DER faults,
variable loading scenarios, and bidirectional fault current.
A wide volume of works documented in the literature are based on local measurements to enhance the reliability of microgrid
protection. Some of them include Compressed sensing and regression tree-based fault detection technique [17], superimposed current
component and wigner distribution function based approach [18], Bagging tree based protection algorithm [19], differential current
derivative processed through decision tree (DT) and k-nearest neighbour (kNN) [20], Variable mode decomposition based protection
technique [21,22]. Despite their claim in providing reliable protection, this applicability of these techniques has not been examined
under wide variation in the PV irradiance levels and outage conditions of DERs. As the PV based sources are highly influenced with the
generation uncertainty due to weather intermittency and outage, the protection schemes must be immune to such dynamic conditions.
While addressing the challenges in development of protection scheme, the signal processing techniques play a crucial role in
deriving useful features by capturing the transient profile from the fault events and ignoring redundant information. Recent research
has witnessed a surge in employing digital signal processing techniques like Wavelet transform (WT) [23], Fast Fourier transform (FFT)
[24], Empirical Mode Decomposition (EMD) [25], and Hilbert transform (HT) [26]. While FFT may have limitations in time-frequency
resolution, WT offers finer resolution. The application of Continuous Wavelet Transform (CWT), Discrete Wavelet Transform (DWT),
and Wavelet Packet Transform (WPT) [27] as a feature extraction technique for fault detection primarily involves simulation studies,
which possess limitations in capturing the transients caused post-fault inception, leading to inefficient fault detection. On the other
hand, the SWT has found extensive applications in image processing due to its inherent characteristics [28,29]. The SWT offers several
advantages over DWT, including shift-invariance, better time-frequency localization, reduced boundary effects, adaptability to
non-stationary signals, and ease of interpretation. However, its implementation in fault detection and identification in DC Microgrid
has not been explored so far. Unlike other Wavelet Transform-based techniques, each decomposition level in SWT maintains the same
number of samples for approximation and detail coefficients as the main signal. This feature ensures consistency in signal length,
enhancing the reliability of SWT for noise reduction, peak and valley detection, and real-time signal analysis compared to DWT [30].
Alongside the feature extraction technique, the efficient machine learning approach also play significant role in the prediction of
fault. Though there are numerous widely used machine learning algorithms including kNN, Support Vector Machine (SVM), DT which
have been claimed to provide effective solution with regard to the fault detection and classification in microgrids. However, for the
conditions involving complex datasets, the response of individual classifier is biased. To avoid such biasness towards in prediction,
ensemble classifiers are preferred which combine multiple classifiers to make prediction and are considered more accurate and robust
than the individual classifiers. Among the commonly used ensemble techniques, the Adaboost ensemble classifier operates iteratively,

2
R. Kumar et al. Computers and Electrical Engineering 120 (2024) 109683

capitalizing on the errors of weak classifiers, which makes it suitable for the complex applications. The performance of ensemble
classifier can be further enhanced by optimizing the hyperparameters for a given dataset. In this regard, the hyperparameters of
Adaboost algorithm (number of learners and number of splits) has been tuned using the Grey wolf optimization technique.
In this regard, to exploit the inherent capabilities SWT in capturing the transient event and extract the useful information for the
classifier for prediction of fault, and accurate prediction of the fault events for the given dataset, this study proposes a joint framework
of SWT and OAE based protection technique for DC microgrid.
In the proposed scheme, the post-fault raw voltage and current signals acquired at the DC bus under a wide variation in the
generation level of PV-based DER, fault parameters, and operating conditions, including DER outage and load variation, are processed
using SWT to derive discriminatory attributes. The standard deviation of detailed coefficients obtained constitutes the input dataset for
the OAE-based classifier for performing fault detection and classification. The proposed SWT-OAE based protection scheme has been
extensively validated for different fault scenarios involving a wide variation in fault parameters and other operating conditions. The
proposed scheme has been found to efficiently perform the intended tasks with a high degree of dependability, security, and accuracy,
in addition to improved robustness against weather intermittency and DER outage. The performance of the proposed scheme has also
been compared with the other ensemble-based classifiers including Bagging Tree and Random subspace.
The significant contributions of the developed approach include:

1. Reliable protection framework for DC microgrid based on the single end data utilizing Stationary Wavelet Transform based feature
extraction technique and Optimal Adaboost Ensemble Classifier
2. Exploiting the inherent characteristics of SWT capable of capturing the fault transients to derive discriminatory attributes per­
taining to distinct fault profiles, thereby enhancing the reliability of protection.
3. Implementation of the OAE-based classifier modules with enhanced input-output mapping ability by optimizing the hyper­
parameters of Adaboost classifier through the GWO method and extensive evaluation of its performance under a wide variation in
fault parameters and other operating conditions of DC Microgrid.
4. Proposed scheme has been found to be fault resilient to weather intermittency and DER outage conditions

Fig. 1. The test DC Microgrid system considered under present study.

3
R. Kumar et al. Computers and Electrical Engineering 120 (2024) 109683

5. Comparison of the proposed SWT-OAE scheme with Bagging Tree and Random subspace based ensemble classifiers to examine the
superiority of the proposed scheme in providing suitable protection.

The organization of the paper is as follows: Section 2 illustrates the DC microgrid system considered in this study. Section 3 de­
scribes the development of the proposed protection methodology, while Section 4 presents the performance analysis. Finally, Section 5
includes the concluding remarks on the proposed work.

2. The test DC microgrid system

The schematic representation of the DC microgrid system considered under present study is illustrated in Fig. 1 [31]. The depicted
DC Microgrid system has been simulated under MATLAB/Simulink environment. The DC microgrid configuration consists of several
key components, including photovoltaic (PV) and diesel generator integrated DERs, along with the DC loads, in synchronization with
the AC grid. The PV system comprises a network of multiple PV generators connected in parallel, interfaced with the DC Bus through
DC-DC converters. To facilitate the conversion from DC to AC, Voltage Source Converters (VSCs) are employed, featuring high Pulse
Width Modulation (PWM) switching frequency and Phase-Locked Loop (PLL) control for synchronization with the AC grid.
Under normal operating conditions, the DC cables within the microgrid typically demonstrate resistive behaviour. For the purpose
of protection analysis, the current-voltage signals have been recorded at the DC bus with simulation of diverse fault and no-fault
scenarios. The extensively generated signals serve as the crucial input data for assessing the performance and efficacy of the pro­
posed protection methodologies. The detailed specification of each component of the DC microgrid with their rating is mentioned in
Table 1. The variation of fault and other parameters considered for simulation of scenarios is listed in Table 2.

3. Proposed protection algorithm

3.1. Framework of proposed protection technique

The framework of the hybrid protection algorithm, which exploits the inherent capabilities of the SWT in terms of deriving the
discriminatory attributes and Grey Wolf Optimization (GWO) tuned Adaboost Ensemble classifier for better input output mapping to
impart robustness to the protection approach, can be outlined as follows (Fig. 2):

• Generation of voltage-current signals: The current and voltage signals are generated by carrying out extensive simulation of the DC
microgrid and recorded at the DC bus, with wide variations in fault resistance, fault location, and loading conditions.
• Feature extraction using SWT: The feature dataset is constructed by computing the Root mean square (RMS) values of detailed
coefficients derived from the time-domain signals using the third-level decomposition by SWT.
• Tuning of hyperparameters of Adaboost Ensemble Classifier: Utilizing GWO, the optimal hyperparameters of the Adaboost
ensemble classifier, including the number of learners (l) and number of splits (s), are determined. The optimization process aims to
attain an optimal correspondence between the feature space and classifier output concerning the classification accuracy.
• Training of OAE Classifier: The optimal architecture of the Adaboost ensemble classifier achieved using GWO is trained, and the
classification accuracy is evaluated.
• Prediction by OAE Classifier Module: The OAE module predicts the final output through weighted majority voting, ensuring
resilient fault detection and classification.

Table 1
Specification of system components of DC Microgrid.
Components of DC Microgrid Parameters Specification

PV Array (PV1, PV2, PV3, PV4) 4× 100 kW 400 kW


DG Voltage Rating 480 V
DC Bus Voltage Rating 500 V
PV Cable: L1 0.3 km
(R=0.641 Ω/ km, L2 1.5 km
L=0.34 Mh/ km, L3 0.35 km
C=0.1 µF/ km) L4 0.75 km
L5 0.4 km
L6 0.5 km
L7 0.45 km
L8 0.95 km
DC Load Cable: L9, L10 0.5 km each
(R=0.641 Ω/ km, L11, L12 0.5 km each
L=0.34 Mh/ km,
C=0.1 µF/ km)
Load (DC Motor) Power Rating 20 kW
DC Load (Resistive) Power Rating 50 kW
AC Grid Voltage Rating 25kV

4
R. Kumar et al. Computers and Electrical Engineering 120 (2024) 109683

Table 2
Range of variation of fault and other parameters.
S. No Name of Parameter Specification/ Range of Variation

1 Features Standard deviation of third-level detailed coefficients of Voltage and Current derived using SWT
2 No. of features 2
3 Type of fault Pole-Pole (PP) and Pole-Ground (PG)
4 Fault Resistance 0 - 50 Ω
6 Fault Location Variation in the range of 0.1 to 1 km
7 DERs PV and DG
8 PV Irradiance Variation 100 to 1000 W/m2
9 Hyperparameters of Adaboost Classifier No. of learners and no of splits

Fig. 2. Framework of proposed protection algorithm.

• Generation of Relay Trip Signal: Based on the predicted output from the OAE modules, relay trip signals are generated for the
respective breakers, instigating appropriate protective measures in response to the detected faults.

3.2. Extraction of attributes using SWT

The primary process to enhance the efficacy of a classifier involves feature extraction, which aims to derive the useful and
discriminatory attributes from raw time-domain signals for each class. This process enables the formation of a dataset, or feature set,

5
R. Kumar et al. Computers and Electrical Engineering 120 (2024) 109683

where the distinct characteristics of each class can be captured. Augmented segregation between the data of diverse classes streamlines
the development of a classifier which is less vulnerable to noise and more resistant to overfitting. Through the extraction of pertinent
features, the classifier acquires insights into the unique characteristics of each class, facilitating precise classification and resilient
performance in real-time scenarios.
In this context, the SWT is specifically employed in the present study to derive discriminatory attributes, utilizing the current and
voltage signals recorded at the DC Bus for executing the intended protection tasks, thereby minimizing the computational time thus
making it more suitable for real-time protection applications.
Unlike other Wavelet Transform-based methodologies, each decomposition level in SWT maintains an identical number of samples
for approximation and detail coefficients as the original signal. This aspect ensures uniformity in signal length, augmenting the
dependability of SWT for noise reduction, peak and valley detection, and real-time signal analysis compared to DWT. Furthermore,
SWT exhibits diminished oscillations and noise in measured signals, rendering it advantageous for feature extraction and presenting
lower sensitivity to noise than other Wavelet Transform-based techniques.
To tackle these challenges, the current study employs the SWT algorithm as a feature extraction tool, integrating its transient
capturing capability into the protection approach for the DC microgrid system. The proposed approach offers a promising solution to
enhance the reliability and robustness of the developed relaying algorithm.
The depiction provided in Fig. 3 illustrates a three-level decomposition of signal using Stationary Wavelet Transform. In this SWT
representation, the coefficient output consists of the approximation coefficients of the highest level and the detail coefficients of each
level. Unlike the discrete wavelet transform, the SWT does not entail sub-sampling. Consequently, the output of each level maintains
the same length as the input signal, preserving valuable information throughout the transformation process.
The Root Mean Square (RMS) value of detailed coefficients derived from the SWT can be represented mathematically as follows:
Let coef(j) represent the detailed coefficients obtained at j level decomposition using SWT.
Then, the standard deviation of these detailed coefficients is calculated as:
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
√ N
√1 ∑
RMScoef(j) = √ (xi )2 (1)
N i=1

where:
N is the number of coefficients at level j, xi represents the ith coefficient at level j,
The derived set of attributes using (1) consisting of the standard deviation of the approximate coefficients of signals obtained using
SWT is represented as:
f = [RMSdcI , RMSdcV (2)
The training dataset T, consisting of input samples (y) and feature variables (f) has been formulated as:
T = {(yk , ft ), k = 1, 2, ……n; t = 1, 2} (3)
The feature extraction process is followed by the development of OAE based classification modules for performing the intended
tasks of fault detection and classification.

3.3. Adaboost based ensemble classifier

The Decision Tree classifier has gained considerable attention in the field of data mining, especially for classification tasks. This
popularity stems from its straightforward implementation and reliable performance, particularly with multi-dimensional datasets
[32]. However, when confronted with complex datasets, decision trees may demonstrate suboptimal performance during testing,

Fig. 3. SWT Decomposition upto 3 level.

6
R. Kumar et al. Computers and Electrical Engineering 120 (2024) 109683

primarily due to their tendency to exhibit bias towards specific classes. To address this limitation, an ensemble approach is often
employed, replacing a single decision tree with a collection of trees [33]. In ensemble learning, the final prediction is determined by
aggregating the outputs of multiple decision trees through a voting mechanism. This ensemble strategy helps mitigate the potential
bias introduced by individual trees, enhancing the overall performance and robustness of the classification model.
Among the popular ensemble methods, the Adaboost classifier offers several advantages that contribute to its popularity and
effectiveness in classification tasks including high accuracy obtaining ability by combining and exploiting weak learners, versatility in
handling wide range of classification problems, prone to overfitting, ease of implementation and improved robustness with reduced
sensitivity to noisy data.
These merits associated with Adaboost algorithm makes it popular for solving various complex classification problems.

3.3.1. Optimization of adaboost ensemble classifier


The final output of the Adaboost ensemble classifier model relies on optimally tuned hyperparameters, including the number of
learners (decision trees), and the number of splits. These hyperparameters are not predetermined and usually require iterative
experimentation to identify the optimal values. In this regard, the process of determining the optimal hyperparameters is treated as an
optimization problem. This involves formulating an objective function aimed at minimizing the classification error by tuning the
decision trees optimally. The objective function f(x) is expressed as follows:

∑ N ⃒

f(x) = |Target(i) − Predicted(i)|⃒ (4)
i=1
⃒ →
x= Y

̅→
where, Y = [l, s, ] is the vector containing the number of learners (trees)(l), and the number of splits(s).
By systematically assessing the objective function across various values of x, the most effective configuration of decision trees and
features can be identified, thereby improving the performance of Adaboost model. In the present work, the GWO is employed for fine-
tuning the hyperparameters of the proposed classifier. The subsequent sections elaborate on the fundamentals of the GWO algorithm
and its application in optimizing the classifier’s hyperparameters.

3.3.2. Grey wolf optimization algorithm


The GWO is a metaheuristic optimization algorithm inspired by the social structure and hunting dynamics of grey wolves.
Introduced by Mirjalili et al. [34], GWO emulates the hierarchical organization within a wolf pack, where individuals are categorized
into four roles: alpha, beta, delta, and omega, representing the pack leaders. This algorithm models the hunting strategy, where the
alpha wolf guides the pack, followed by beta, delta, and omega members. For a detailed overview of the GWO methodology, please
refer to [34]. GWO has proven its efficacy across various optimization tasks, showcasing its ability to discover near-optimal solutions
within complex search spaces [35–38]. Given its success, this study leverages GWO for hyperparameter optimization of the bagging
tree classifier.

3.3.3. Implementation of GWO for hyperparameter tuning of adaboost ensemble classifier


To implement the GWO for hyperparameter tuning of the bagging tree classifier, we first randomly generate the initial population
within specified search spaces for the hyperparameters. For instance, the number of learners (decision trees) is initialized in the range
of [1, 100], the number of splits falls within [1,20].
Following the initialization, the GWO algorithm iterates through a fixed number of iterations, during which it optimizes the
hyperparameters by evaluating the objective function defined in Eq. (4).
Throughout the iterations, the algorithm updates the positions of the alpha, beta, delta, and omega wolves according to their
hunting behaviour, converging towards the optimal solution. Once the predetermined number of iterations is reached, the algorithm
halts, and the best hyperparameters for the problem are obtained based on the positions of the wolves. These hyperparameters
represent the configuration that maximizes the performance of the Adaboost classifier within the given search space. The pseudo code
illustrating the implementation of GWO for hyperparameter tuning of Adaboost Ensemble classifier is given below:

Pseudo code for hyperparameter tuning of Adaboost Ensemble Classifier using GWO

Input: Training dataset containing the RMS value of detailed coefficients derived using third- level decomposition of time-domain voltage and current signals
through SWT
Output: Optimal set of hyperparameters of Adaboost ensemble classifier
1. Define initial parameters of GWO algorithm and set search space for hyperparameters of Adaboost ensemble classifier.
2. Initialize the population of grey wolves randomly.
( )
Wk+1
i,j = Wki,j + Rki,j Wmax
j − Wmin
j (5)
where, Wk+1
i,j is the jth grey wolf for ith population in (k+1)th iteration and Wki,j is the jth grey wolf for ith population in kth iteration. Rki,j is any random number
between 0 and 1 which are evenly distributed. “max” and “min” are the maximum and minimum values of the decision variable, respectively.
3. Initialize the number of learners (l) and number of splits (s) within their respective search space.
4. Evaluate the fitness of each wolf based on the following objective function defined in (Eq. (4)) and identify alpha(Wα), beta(Wβ), delta(Wδ), and omega(Wω).
∑N
f(x) = i=1
|Target(i) − Predicted(i)|| →
x= Y
5. Update the position of each wolves based on their hierarchy and hunting characteristics.
6. If maximum number of iterations has been achieved, obtain the optimal set of hyperparameters else go to step 4.

7
R. Kumar et al. Computers and Electrical Engineering 120 (2024) 109683

For simulations, a population size of 50 and maximum number of iterations is set 100. After 33 independent simulations, the
optimal hyperparameters of Adaboost ensemble classifier have been obtained as:

1. Number of learners - 47
2. Number of splits- 9

The convergence characteristics of GWO algorithm applied for hyperparameter tuning of Adaboost Ensemble classifier is illustrated
in Fig. 4. From the figure, it can be seen that the final value of objective function (classification error) is nearly equal to zero. The
aforementioned optimal hyperparameters of bagging tree classifier have been used in the present study.

4. Performance evaluation

The efficacy of proposed protection algorithm has been examined under diverse fault scenarios. A dependable protection system
should accurately generate the trip signals in the event of fault and should not be influenced by the factors involving DER penetration
level, weather intermittency, and outage conditions. The simulated scenarios included various fault conditions like pole-pole (PP) and
pole-ground (PG), along with considerations for the impact of weather intermittency on Solar PV generators and sudden DER outages.
In addition to this, the performance of proposed scheme has also been assessed under different levels of distributed generation
penetration.
In order to evaluate the proposed scheme’s performance, the testing dataset consisted of 420 fault test cases (210 test cases for each
of the PG and PP faults) and 120 no-fault test cases, encompassing wide variations in parameters. The fault types ’PP’ and ’PG’ were
assigned target labels of ’1′ and ’2′, respectively, with ’0′ representing the no-fault case.

4.1. Performance of SWT-OAE based proposed fault detector and classifier

In this subsection, the effectiveness of proposed SWT-OAE based fault detection/classification module has been examined in terms
of accurately identifying and categorizing faults, and distinguishing between pole-pole (PP) and pole-ground (PG) fault types. As
outlined earlier in Section 3, the proposed protection approach involves utilizing SWT to extract distinguishing features, involving the
extraction of RMS value of third-level detailed coefficients. These extracted features are then fed into the OAE-based classification
module. As a representative case, the current signal recorded at DC Bus with corresponding third-level detailed coefficients derived
using SWT for PP fault inception at t=2.5 s is illustrated in Fig. 5(a) and (b) and the response of proposed scheme with trip signal
generation post 2 ms of fault inception is depicted in Fig. 5(c). Similarly, the post fault behaviour for PG fault along with the response of
proposed scheme is also demonstrated in Fig. 6. As observed from both Figs. 5 and 6, the prompt response of proposed scheme in
isolating the faulty feeder shows the fast relaying speed of the algorithm. In order to demonstrate the voltage profile and corresponding
variation in the feature derived using SWT, the voltage waveform recorded at the DC bus due to PG fault inception at t = 2.5 s is

Fig. 4. Convergence characteristics of GWO.

8
R. Kumar et al. Computers and Electrical Engineering 120 (2024) 109683

Fig. 5. (a) Current signal recorded at DC Bus with corresponding (b) Third-level detailed coefficients derived using SWT for PP fault inception at t =
2.5 s and (c) Trip signal generation post 2 ms of fault inception.

demonstrated in Fig. 7(a) and (b). As observed the significant variation in feature can be seen even for less variation in the voltage due
to impact of fault.
The effectiveness of proposed OAE classifier is demonstrated through the confusion matrix depicted in Fig. 8, revealing an overall
testing accuracy of 99.44 %. The higher degree of accuracy achieved by proposed scheme confirms its capability in effectively
detecting and classifying fault events in DC Microgrid, even under varying operational conditions such as uncertainties in PV-based
DERs and reconfigurations due to DER outages.
To further examine the performance of the proposed OAE classifier in detecting and classifying faults, a Receiver Operating
Characteristic (ROC) curve has been derived, as depicted in Fig. 9. The ROC curve provides a visual representation of the classification
model’s effectiveness, illustrating the trade-off between sensitivity and specificity as the decision threshold changes. With the area
under the curve (AUC) close to 1, as depicted in Fig. 9, the proposed scheme demonstrates its suitability in providing effective pro­
tection to the DC Microgrids.
The effectiveness of the proposed protection algorithm was evaluated in terms of the reliability indices, including dependability
and security, which is the figure of merit for any protection system, is presented in Fig. 10. In order to demonstrate the performance of
proposed OAE based classifier, the comparative analysis has been carried out with Decision Tree and Random subspace-based
ensemble classifiers, as summarized in Table 3. In order to illustrate the comparative results, the bar chart showing the

9
R. Kumar et al. Computers and Electrical Engineering 120 (2024) 109683

Fig. 6. (a) Current signal recorded at DC Bus with corresponding (b) Third-level detailed coefficients derived using SWT for PG fault inception at t =
2.5 s and (c) Trip signal generation post 2.2 ms of fault inception.

performance comparison in terms of reliability indices i.e. dependability, security and accuracy has been depicted in Fig. 11. As
observed, the higher testing accuracy achieved by the proposed OAE-based scheme clearly indicates its outperforming performance as
compared to other ensemble classifiers.
The qualitative comparison of proposed protection scheme has also been carried out other recent existing protection schemes
documented in the literature related to DC microgrid in Table 4. As observed from the comparison, most of the protection schemes
have not considered the two important aspects of protection involving the PV intermittency and DER outage. Though some of the
techniques have considered PV intermittency but in the limited range. So,

4.2. Performance under DER outage

In scenarios where microgrids experience the disconnection of Distributed energy resources (DERs) or outage conditions, the
conventional relays may face challenges due to variations in fault current magnitude and direction. To address this, the performance of
proposed protection scheme has also been evaluated under DER outage scenarios followed by fault occurrences. For instance, Fig. 11
illustrates a scenario where there’s a PV1 outage at t = 2.0 s followed by a pole-ground (PG) fault at t = 2.5 s. The current waveform
recorded at the DC bus is depicted in Fig. 12(a) alongside the corresponding third-level detailed coefficient derived using SWT in
Fig. 12(b). The response of proposed scheme with generation of trip signal generated post 3 ms of the fault inception to the breaker is
depicted in Fig. 12(c). It’s evident that while there’s no significant impact of the PV outage on the derived coefficient plot, there’s a
noticeable variation due to the PG fault at t = 2.5 s. This highlights the effectiveness of our proposed scheme even under DER outage
conditions, showcasing its immunity to such events.

10
R. Kumar et al. Computers and Electrical Engineering 120 (2024) 109683

Fig. 7. (a) Voltage signal recorded at DC Bus with corresponding (b) Third-level detailed coefficients derived using SWT for PG fault inception at t =
2.5 s.

Fig. 8. Performance of proposed OAE based classifier for Predicted Fault (PG and PP) and No-fault classes.

4.3. Performance under weather intermittency of PV based DERs

The microgrids, particularly those incorporating Photovoltaic (PV) based-DERs, often contend with the intermittent operation due
to the inherent variability of irradiance levels directly influencing the generation levels of on-site PV installations. This variability,
influenced by the factors involving geographical location and weather conditions, poses a challenge during fault occurrences in the
distribution line as it may cause the variation in the level of fault current. The intermittent behaviour of PV systems, driven by un­
certain fluctuations in irradiance levels, directly affects the subsequent changes in fault current magnitude, even under similar fault

11
R. Kumar et al. Computers and Electrical Engineering 120 (2024) 109683

Fig. 9. ROC curve of proposed OAE classifier for the considered test cases

Fig. 10. Performance of proposed OAE based Fault Detector and Classifier Module in terms of the reliability indices.

conditions. The traditional protection methods relying on predetermined threshold settings may fail to detect faults amidst the un­
certainty in generation inherent in renewable PV-based DERs.
To enhance the robustness of the protection scheme against generation uncertainty in PV-based DERs, the performance of the
proposed protection algorithm has been extensively evaluated considering fault scenarios involving a wide range of PV irradiance
levels, ranging from 100 to 1000 W/m2. As a representative case, a scenario has been simulated involving a drop in PV irradiance from
1000 to 300 W/m2 at t = 2.0 s, followed by the inception of a PP fault at t = 2.5 s, as depicted in Fig. 13(a). The corresponding third-
level detailed coefficient derived using SWT is illustrated in Fig. 13(b). Notably, there is no impact of the variation in irradiance level
observed, whereas significant variation can be seen at the fault instant in the feature plot at t = 2.5 s. Also, the generation of trip signal

12
R. Kumar et al. Computers and Electrical Engineering 120 (2024) 109683

Table 3
Performance comparison of the proposed OAE based fault detector/classifier module with Bagging Tree and Random Subspace classifiers.
Technique No. of Test scenarios Type of Correctly predicted Correctly predicted No- Reliability Indices (%)
considered Scenario fault scenarios fault scenarios
Dependability Security Accuracy

Proposed OAE 210 PP Fault 210 – 99.52 % 99.16 % 99.44 %


Classifier
​ 210 PG Fault 208 –
​ 120 No Fault – 119
Bagging Tree 210 PP Fault 197 – 93.10 % 77.58 % 89.81 %
Classifier
​ 210 PG Fault 194 –
​ 120 No Fault – 94
Random Subspace 210 PP Fault 191 – 89.52 % 74.60 % 86.48 %
classifier
​ 210 PG Fault 185 –
​ 120 No Fault – 91

Fig. 11. Comparative analysis of the proposed OAE based protection module with Bagging Tree and Random subspace classifiers in terms of
reliability indices.

post 2.4 ms of the fault inception shows the prompt relaying response of proposed scheme. This unequivocally confirms the effec­
tiveness of the proposed scheme in providing suitable protection to the DC microgrid with immunity against weather intermittency of
PV-based renewable DERs.

5. Conclusion

The proposed work entails the development of a resilient hybrid protection framework designed to detect and classify faults in DC
microgrids, effectively addressing challenges such as intermittency in PV-based DERs and outage conditions. By exploiting the inherent
capabilities of the SWT for feature extraction, the developed approach involves capturing the distinct fault profiles with precision. The
fine-tuning of OAE classifier hyperparameters using GWO enhances the algorithm’s input-output mapping ability, ensuring optimal

13
R. Kumar et al.
Table 4
Comparison of the proposed method with other existing methods.
Techniques Basis of Proposed Protection Algorithm [17] [18] [19] [20] [22]
Comparison

DC Microgrid System DC Microgrid test system LVDC Microgrid IEC Microgrid 340 V DC Microgrid LVDC Microgrid LVDC Microgrid
Architecture Radial configuration Radial configuration Radial and Loop Configuration Loop Configuration Radial Configuration Loop Configuration
Protection technique used Stationary Wavelet Transform based feature Regression Tree and Superimposed current Bagging Tree based Differential current Variable mode
extraction technique and Optimal Adaboost Long Short-term tree component and Wigner protection approach derivative with DT and decomposition
14

Ensemble Classifier model distribution function kNN


Type of DERs PV and Synchronous Generator PV, Battery, Flywheel PV, Wind turbine, Synchronous PV, Wind turbine, PV, Wind turbine, PV, Fuel Cell,
Generator, Battery Synchronous Generator, Battery Battery
Battery
Intermittency issues of PV Considered in the range of 100–1000 W/m2 Not considered Not considered Considered Considered Not considered
DER Outage conditions Not considered Not considered Not considered Not considered Not considered Not considered
Local information or One-end local information based Local Information based Local Information based Local Information based Two-end data required Local Information
Communication based
based

Computers and Electrical Engineering 120 (2024) 109683


R. Kumar et al. Computers and Electrical Engineering 120 (2024) 109683

Fig. 12. PV1 outage at t = 2.0 s followed by inception of PG fault at t = 2.5 s (a) Current signal recorded at DC Bus with corresponding (b) third-
level detailed coefficients derived using SWT and (c) the response of proposed scheme with trip signal generation post 3 ms of fault inception.

performance across varying operating conditions. Extensive simulation of scenarios has been conducted to analyse and compare the
performance of proposed scheme with other ensemble techniques. The higher overall accuracy of 99.44 % and prompt relaying speed
(<2.5 ms) achieved by proposed algorithm demonstrate the superiority and efficacy of the proposed protection approach in accurately
detecting and classifying faults as compared to the Bagging and Random subspace-based ensemble classifiers. Furthermore, the
proposed scheme has been rigorously examined against the scenarios considering the impact of PV intermittency and DER outages and
has been proved to be outperforming as compared to other existing techniques on DC microgrid protection, showcasing its resilience
and robustness in such dynamic conditions.

Author agreement statement

We the authors declare that this manuscript is original, has not been published before and is not currently being considered for
publication elsewhere. We confirm that the manuscript has been read and approved by all named authors. We further confirm that the
order of authors listed in the manuscript has been approved by all of us.
We understand that the Corresponding Author is the sole contact for the Editorial process. He/she is responsible for communicating
with the other authors about progress, submissions of revisions and final approval of proofs.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to
influence the work reported in this paper.

15
R. Kumar et al. Computers and Electrical Engineering 120 (2024) 109683

Fig. 13. PV irradiance drop from 1000 to 300 W/m2 at t = 2.0 s followed by inception of PP fault at t = 2.5 s (a) Current signal recorded at DC Bus
with corresponding (b) third-level detailed coefficients derived using SWT and (c) the response of proposed scheme with trip signal generation post
2.4 ms of fault inception.

Data availability

Data will be made available on request.

References

[1] Parhizi S, Lotfi H, Khodaei A, Bahramirad AS. State of the art in research on microgrids: a review. IEEE Access 2015;3:890–925.
[2] Saldarriaga-Zuluaga SD, Lopez-Lezama JM, Mũnoz-Galeano N. Protection coordination in microgrids: current weaknesses, available solutions and future
challenges. IEEE Lat Am Trans 2020;18(10):1715–23.
[3] Dhar S, Patnaik RK, Dash PK. Fault detection and location of photovoltaic based DC microgrid using differential protection strategy. IEEE Trans Smart Grid
2018;9(5):4303–12.
[4] Bacha S, Picault D, Burger B, Etxeberria-Otadui I, Martins J. Photovoltaics in microgrids: an overview of grid integration and energy management aspects. IEEE
Ind Electron Mag. 2015;9(1):33–46. March.
[5] Ullah S, Haidar AMA, Zen H. Assessment of technical and financial benefits of AC and DC microgrids based on solar photovoltaic. Electr Eng 2020;102:
1297–310.
[6] Beheshtaein S, Member S, Cuzner RM, Member S. DC microgrid protection : a comprehensive review. IEEE J Emerg Sel Top Power Electron 2019:1. March.
[7] Joshua AM, Vittal KP. Incremental transient power-based protection scheme for a DC microgrid. Electr Eng 2022;104(4):2281–92.
[8] Krismanto AU, Mithulananthan N, Kamwa I. Oscillatory stability assessment of microgrid in autonomous operation with uncertainties. IET Renew Power Gener
2018;12(4):494–504.
[9] Soleimanisardoo A, Karegar HK, Zeineldin HH. Differential frequency protection scheme based on off-nominal frequency injections for inverter-based islanded
microgrids. IEEE Trans Smart Grid 2018;10(2):2107–14.
[10] Hooshyar A, Iravani R. A new directional element for microgrid protection. IEEE Trans Smart Grid 2017;9(6):6862–76.
[11] Saber A, Zeineldin HH, EL-Fouly THM, Al-Durra A. A signed correlation index- based differential protection scheme for inverter-based islanded microgrids. Int J
Electr Power Energy Syst 2023;145:108721.
[12] Manditereza PT, Bansal RC. Protection of microgrids using voltage-based power differential and sensitivity analysis. Int J Electr Power Energy Syst 2020;118:
105756.
[13] Jia K, Li Y, Fang Y, Zheng L, Bi T, Yang Q. Transient current similarity based protection for wind farm transmission lines. Appl Energy 2018;225:42–51.
[14] Azizi S., Sanaye-pasand M., Abedini M., and Hasani A., “A traveling-wave-based methodology for wide-area fault location in multiterminal dc systems,” vol. 29,
no. 6, pp. 2552–60, 2014.
[15] Meghwani A., Srivastava S.C., and Chakrabarti S., “A new protection scheme for DC microgrid using line current derivative,” pp. 1–5, 2015.

16
R. Kumar et al. Computers and Electrical Engineering 120 (2024) 109683

[16] Naik J., Dhar S., and Dash P.K., “Effective fault diagnosis and distance calculation for photovoltaic-based DC microgrid using adaptive EWT and kernel random
vector functional link network,” 2019.
[17] Salehimehr S, Mahdi Miraftabzadeh S, Brenna M. A novel machine learning-based approach for fault detection and location in low-voltage DC microgrids.
Sustainability 2024;16(7):2821.
[18] Waqar H, Bukhari SBA, Wadood A, Albalawi H, Mehmood KK. Fault identification, classification, and localization in microgrids using superimposed components
and Wigner distribution function. Front Energy Res 2024;12:1379475. https://doi.org/10.3389/fenrg.2024.1379475.
[19] Tiwari SP, Koley E. A local measurement-based protection scheme for DER integrated DC microgrid using bagging tree. Iran J Electr Electron Eng 2022;18(4).
[20] Saxena A, Sharma NK, Samantaray SR. An enhanced differential protection scheme for LVDC microgrid. IEEE J Emerg Sel Top Power Electron 2022;10(2):
2114–25.
[21] Pan P, Mandal RK. Fault detection and classification in DC microgrid clusters. Eng Res Express 2023;5(2):025010.
[22] Sharma NK, Samantaray SR, Bhende CN. VMD-enabled current-based fast fault detection scheme for DC microgrid. IEEE Syst J 2021;16(1):933–44.
[23] Manohar M, Koley E, Ghosh S. Microgrid protection under wind speed intermittency using extreme learning machine. Comput Electr Eng 2018;72:369–82. Nov.
[24] Netsanet S, Zhang J, Zheng D. Bagged decision trees based scheme of microgrid protection using windowed fast fourier and wavelet transforms. Electronics
2018;7(5) (Basel)61 May.
[25] Sarangi S, Biswal C, Sahu BK, Samanta IS, Rout PK. Fault detection technique using time-varying filter-EMD and differential-CUSUM for LVDC microgrid system.
Electric Power Syst Res 2023;219:109254. June.
[26] Shaik M, Shaik AG, Yadav SK. Hilbert–Huang transform and decision tree based islanding and fault recognition in renewable energy penetrated distribution
system. Sustain Energy Grids Netw 2022;30:100606. June.
[27] Saleh SA, Ahshan R, Abu-Khaizaran MS, Alsayid B, Rahman MA. Implementing and testing $ d $–$ q $ WPT-based digital protection for microgrid systems. IEEE
Trans Ind Appl 2013;50(3):2173–85.
[28] Zifan A, Hassan Moradi M, Gharibzadeh S. Microarray image enhancement by denoising using decimated and undecimated multiwavelet transforms. Signal
Image Video Process 2010;4:177–85.
[29] Wang XY, Yang HY, Fu ZK. A new wavelet-based image denoising using undecimated discrete wavelet transform and least squares support vector machine.
Expert Syst Appl 2010;37(10):7040–9.
[30] Yılmaz A, Küçüker A, Bayrak G, Ertekin D, Shafie-Khah M, Guerrero JM. An improved automated PQD classification method for distributed generators with
hybrid SVM-based approach using un-decimated wavelet transform. Int J Electr Power Energy Syst 2022;136:107763.
[31] Dhar S., Member S., Patnaik R.K., Dash P.K., and Member S., “Fault detection and location of photovoltaic based DC microgrid using differential protection
strategy,” vol. 3053, no. c, 2017.
[32] Netsanet S., Zhang J., and Zheng D., “Bagged decision trees based scheme of microgrid protection using windowed fast fourier and wavelet transforms,” no.
March, pp. 1–25, 2018.
[33] Samantaray SR. Ensemble decision trees for high impedance fault detection in power distribution network. Int J Electr Power Energy Syst 2012;43(1):1048–55.
[34] Mirjalili S, Mohammad Mirjalili S, Lewis A. Grey wolf optimizer. Adv Eng Softw 2014;69:46–61.
[35] Faris H, Aljarah I, Al-Betar MA, Mirjalili S. Grey wolf optimizer: a review of recent variants and applications. Neural Comput Appl 2018;30:413–35.
[36] Al-Tashi Q, Rais HMd, Abdulkadir SJ, Mirjalili S, Alhussian H. A review of grey wolf optimizer-based feature selection methods for classification. Evol Mach
Learn Tech Algorithms Appl 2020:273–86.
[37] Makhadmeh SN, Al-Betar MA, Doush IA, Awadallah MA, Kassaymeh S, Mirjalili S, Zitar RA. Recent advances in grey wolf optimizer, its versions and
applications. IEEE Access 2023.
[38] Long W, Jiao J, Liang X, Tang M. An exploration-enhanced grey wolf optimizer to solve high-dimensional numerical optimization. Eng Appl Artif Intell 2018;68:
63–80.

17

You might also like