Engineering Applications of Artificial Intelligence 129 (2024) 107516

Contents lists available at ScienceDirect

Engineering Applications of Artificial Intelligence

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

An efficient power extraction using artificial intelligence based machine

learning model for SPV array reconfiguration in solar industries
Mona Sharma a, Smita Pareek b, *, Kulwant Singh a, c, **
a
Department of Electronics & Communication Engineering, Manipal University Jaipur, 303007, Rajasthan, India
b
Department of Electrical Engineering, B K Birla Institute of Engineering & Technology, Pilani, 333031, Rajasthan, India
c
FlexMEMS Research Centre (FMRC), Manipal University Jaipur, Jaipur, 302007, Rajasthan, India

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

Keywords: In the field of solar industries, extracting maximum power from solar photovoltaic systems under partial shading
Total cross-tied conditions has gained significant attention in recent years. One of the most efficient concepts brought out to
Sudoku extract the power output of the existing solar photovoltaic system is the reconfigurable solar photovoltaic system.
Solar photovoltaic
Numerous static and dynamic reconfiguration techniques are mentioned in the literature. This work proposed a
Maximum power
Partial shading losses
Modified Sudoku reconfiguration based on static techniques and compared to the most common existing
Artificial intelligence configuration i.e., Total-Cross-Tied and Sudoku photovoltaic array configurations. In addition, an Artificial
Fuzzy Expert System Intelligence-based machine learning model (Fuzzy Expert System) is implemented for the prediction of suitable
configurations for solar photovoltaic arrays under partial shading conditions. The performance of the FES model
is evaluated by comparing the predicted results and the results obtained from the simulations on 5 × 4, 6 × 4, 6
× 6 and 9 × 9 PV arrays. Results demonstrated that the implemented FES model generated accurate results and
0% MAPE in all 33 sample cases for predicting the best suitable configuration. This realistic, simple, and cost-
effective fuzzy model can be utilized to replace existing estimation systems that employ the use of complex
technological analysis and simulation models. Thus, this approach can be very effectively used in the solar in­
dustry for selecting the configuration for installing solar panels.

1. Introduction 1.2. Partial shading (PS) effects

1.1. Solar photovoltaic system and various factors that impact its PSC is a most common reason for light restriction on solar array that
performance can result in power loss of PV systems. The shading over the solar arrays
blocks the solar radiation and reduces complete absorption, which leads
Solar industries encounter significant challenges to wider imple­ reduction of energy received by the individual solar cells. This condition
mentation and acceptance despite technological development and turns out to be difficult as it reduces the maximum power of the PV panel
numerous benefits. The reason is that the output efficiency of the Solar and thus reduces the consistency of this technology and in turn the
photovoltaic (SPV) system is restricted by several factors, including low reliability of this technology. Researchers have focused on establishing
energy conversion, variable irradiation, atmospheric conditions, variation methods for extracting the maximum power to overcome the effects of
in temperature, panel layout, reflection, partial shading, and so on (Par­ PS and to increase the amount of electricity generated (Sharma et al.,
eek and Dahiya, 2016). The most significant cause of power loss among all 2020; Khuran and Pareek, 2022). Various PS loss reduction techniques
is the non-uniform irradiation that solar panels receive due to partial are covered in next section.
shading conditions (PSCs) which leads in annual power losses of 3–6% of
the installed capacity in large PV systems (Kumar and Kumar, 2017).

* Corresponding author.
** Corresponding author. Department of Electronics & Communication Engineering, Manipal University Jaipur, 303007, Rajasthan, India.
E-mail addresses: smitapareek.bkbiet@gmail.com (S. Pareek), eckulwant@gmail.com (K. Singh).

https://doi.org/10.1016/j.engappai.2023.107516
Received 9 March 2023; Received in revised form 9 November 2023; Accepted 13 November 2023
Available online 6 December 2023
0952-1976/© 2023 Elsevier Ltd. All rights reserved.
M. Sharma et al. Engineering Applications of Artificial Intelligence 129 (2024) 107516

Nomenclatures
AI Artificial intelligence FL Fuzzy logic
ANFIS Adaptive-neuro fuzzy inference systems MAPE Mean absolute percentage error
BP Back-propagation ML Mismatch losses
PV Photovoltaic systems PM Maximum power
ML Machine learning VM Voltage at Maximum Power
TCT Total cross tied IM Current at Maximum Power
SP Series-parallel GMPP Global maximum power point
HC Honeycomb PER Performance enhancement ratio
FES Fuzzy expert system WT Wavelet transform
PSC Partial shading condition DT Decision tree
MPPT Maximum power point tracking LMPP Local maximum power points
GA Genetic algorithm SPV Solar photovoltaic
NN Neural networks MF Membership function
A Ampere V Voltage
W Watt S Series
ES Expert system P–V Power-voltage
BL Bridge link I–V Current -voltage
WS Without shading LBC Left bottom corner
SN Short narrow RTC Right Bottom corner
LN Long narrow DN Diagonal
LW Long wide LS L-shape
SR Single row MS Multi storied
OB-1 Oblique-1 CN Centre
OB-2 Oblique-2 OB-3 Oblique-3

1.3. Partial shading loss reduction techniques control circuits, micro-converters, multi-level converters, compensa­
tion, battery equalizer, energy recovery, etc. The implementation of
PS losses could be minimized via passive or active methods. Active these techniques for high power ratings solar plants is challenging since
techniques use active elements like solid-state switches, whereas passive each plant requires many converters, which raises the costs, which is a
techniques use passive elements like bypass diodes. Connecting bypass common drawback, thus makes these techniques less popular (Richard
diodes across each PV panel is common passive approach reported in et al., 2011; Giovanni et al., 2015; Hyok et al., 2010; Villa et al., 2013).
literature. However, these diodes limit the array from generating the The effective concept brought out to mitigate the drawbacks of
maximum amount of power when partially shaded and introduce many above-mentioned techniques is the reconfigurable solar photovoltaic
local maximum power point (LMPP) in power-voltage characteristics. system which is described in next section. Table 1 demonstrates a
Several Maximum Power Point Tracking (MPPT) approaches and algo­ summarized review of various PS loss reduction techniques which are
rithms have been presented in the literature to identify global maximum reported in the literature.
power point (GMPP). However, selection from the available solutions
becomes difficult because these systems differ in relations of depend­
1.4. Review of various PV array reconfiguration techniques
ability, price, tracking speed, category and number of sensors needed,
hardware necessities etc. (Gosumbonggot and Fujita, 2019; Nivedha and
1.4.1. Basic interconnection schemes
Vijayalaxmi, 2021). Other techniques based on active devices reported
The literature suggests several PV array interconnection methods to
in the literature to deal with PS losses comprise of AC panels, generation
address PSCs issues like SP (Series-Parallel), BL (Bridge-Linked), TCT

Table 1
Summarized review of different PS loss reduction techniques available in the literature.
Technique Goal/objective Advantages Disadvantages References

Bypass Diode Provide a bypass for current flow Protect against hotspots Limit the maximum attainable (Pareek and Dahiya, 2016; Kumar and Kumar, 2017)
when cell is reverse biased. and damage of solar cell. power. Introduce many LMPP.
Non optimal operation.
Control Circuits To achieve optimum power, alter Optimal power, Require many sensors and (Richard et al., 2011; Giovanni et al., 2015; Hyok et al.,
the control circuits between the PV increased efficiency, and switches and therefore 2010; Villa et al., 2013)
panels in real time. high reliability. relatively expensive.
MPPT Track maximum power point of Increased running time Increased overall cost due to (Gosumbonggot and Fujita, 2019; Nivedha and
panel/array and operates it at and efficiency. dependency on switches and Vijayalaxmi, 2021)
optimal point. sensors.
Basic Panels are connected in series/ Reduced mismatch Fail to find optimal solution (Das et al., 2017; Verma et al., 2019; Winston et al.,
Interconnection parallel to meet the voltage and losses, improve under variable shading. 2020)
Schemes power specifications. efficiency, and simple.
Static configuration Group the panels based on shade Simple and economical Fail to find optimal solution (Krishna and Moger, 2019; Vijayalekshmy et al., 2016;
dispersion to increase the power and do not require under variable shading. Reddy and Yammani, 2021; Dhanalakshmi and
output. sensors or switches. Rajasekar, 2018a, 2018b; Nasiruddin et al., 2019)
Dynamic Interconnection switching Stable, reliable, and Complicated hardware and (Guillermo et al., 2005; Jazayeri et al., 2017; Rajan
configuration technique based on optimization capable of identifying complex optimization et al., 2017; Babu et al., 2015; Fathy, 2018; Yousri
algorithms to produce more power. the optimum solution. techniques. et al., 2020a, 2020b; Dalia et al., 2020; Yang et al.,
2021; Ahm et al., 2020; Bayoumi et al., 2022; Zhang
et al., 2021a, 2021b, 2022)
Artificial Useful for SPV system modeling, Fast learning, reliable Handle only trained problems. (Yap et al., 2020; Gupta et al., 2022; Machina et al.,
Intelligence prediction, simulation, cost effective, accurate 2022; Mellit et al., 2008; Wu et al., 2019; Kouaissah
optimization, and control. and fault tolerant. and Hocine, 2020; Sridharan, 2021)

2
M. Sharma et al. Engineering Applications of Artificial Intelligence 129 (2024) 107516

(Total-Cross-Tied), and HC (Honeycomb). In PSC, the TCT performed Reddy and Yammani, 2021). Additional puzzle-based technique re­
better in comparison to other interconnection schemes (Das et al., 2017; ported was the dominant square technique and implemented on 5 × 5
Verma et al., 2019). However, in TCT extra ties are significant, therefore and 9 × 9 PV arrays (Dhanalakshmi and Rajasekar, 2018a). To mitigate
wire loss, and cost due to extra ties is significant (Winston et al., 2020). partial shading losses in an PV array, an odd-even reconfiguration
strategy was also reported on a 4 × 4 PV array (Nasiruddin et al., 2019).
1.4.2. Dynamic array reconfiguration approaches Competence square approaches, on a 9 × 9 PV array (Dhanalakshmi and
To address the issue present in existing interconnection schemes, Rajasekar, 2018b), is also one of the most recent puzzle-based method.
various optimization based dynamic reconfiguration techniques have Overall, it is worth underlying that static reconfiguration techniques are
been documented in literature. These techniques are based on shade less expensive approach because these techniques require a complex
dispersion achieved by altering electrical connections of PV array hardware setup which leads to an unrealistic number of sensors and
without changing physical location. In recent year various dynamic switches and increases additional cost. Numerous static and dynamic
array reconfiguration such as Electrical array reconfiguration (EAR) reconfiguration techniques are available for PV array. It might be
(Guillermo et al., 2005), Irradiation equalization (IE) reconfiguration difficult to choose the best method out of all those that are available for a
and Adaptive array reconfiguration (AAR) (Jazayeri et al., 2017) are specific requirement. Additionally, these methods needed an extensive
reported in the literature based on switching between the connections of knowledge of PV technology. Thus, using AI to address these issues has
a solar PV array to mitigate partial shading losses. In addition, various been identified as a possible alternative.
meta-heuristic optimization algorithms including, standard deviation
genetic algorithm (10 × 10, 9 × 9), particle swarm optimization (9 × 9), 1.5. Artificial intelligence techniques implemented in SPV system
grasshopper optimization algorithm (9 × 9), harris hawks optimizer
(HHO) and modified harris hawks optimizer (9 × 9, 6 × 4, 6 × 20) based For modeling, prediction, simulation, optimization, and control of
on scheme of prey catching by hawks, grey wolf optimization algorithm SPV systems, AI approaches have been used in a variety of applications.
(GWOA) (Rajan et al., 2017; Babu et al., 2015; Fathy, 2018; Yousri et al., AI-based techniques have the advantages of providing an alternative to
2020a; Dalia et al., 2020), artificial ecosystem optimizer (9 × 9, 6 × 20, conventional physical modeling techniques, requiring less computa­
6 × 20), flow regime algorithm (9 × 9), coyote optimization algorithm tional effort, requiring no knowledge of internal system characteristics,
(COA), marine predators algorithm (9 × 9, 16 × 16, 25 × 25), are a and providing a compact solution for multivariable problems. Literature
significant addition in this research area (Yousri et al., 2020b; Yang validates the potential of AI as a design tool for SPV systems. AI-based
et al., 2021; Ahm et al., 2020; Bayoumi et al., 2022). An innovative machine learning model is involved in the design of SPV systems in all
technique optimal mileage-based PV array reconfiguration (OMAR) is areas, including control and monitoring. AI consists of several tech­
developed to increase the power output of a PV power plant and reduce niques, specifically, expert system (ES), artificial neural network (ANN),
the additional capacity and mileage costs brought on by the power genetic algorithm (GA), fuzzy logic (FL), machine learning (ML), various
fluctuation in a performance-based frequency regulation market. This optimization techniques, decision tree, wavelet transform etc. (Yap
technique was implemented on 10 × 10 SPV array (Zhang et al., 2021a). et al., 2020; Gupta et al., 2022; Machina et al., 2022; Mellit et al., 2008).
Current research has also contributed to the development of an evolu­ Implementation of AI techniques in various areas of SPV system is shown
tionary based pareto optimization algorithms for bi-objective optimi­ in Fig. 1. Fuzzy expert system (FES) is a common choice among the AI
zation method for photovoltaic array reconfiguration, which aims to techniques that are available for use in solar industries research because
simultaneously maximize output power and reduce number of switches. of the simple model, adaptability, and ease. FES enable machines to
The generation efficiency can be increased while the switching control make decision by analyzing data and selecting one among several op­
complexity is significantly reduced, and the operating life of the tions. FES is based on a rule-based inference method which is
switching devices can be extended by this method (Zhang et al., 2021b). pre-established rule systems from strong knowledge base. Fuzzy logic is
A novel efficient multi-agent negotiation algorithm with an auctioneer proved as significant technique in the solar energy industries for pre­
and multiple bids is developed in literature with the multi-round dicting and controlling the performance of the energy conversion system
negotiation and the random re-initialization and implemented on a (Wu et al., 2019; Kouaissah and Hocine, 2020; Sridharan, 2021).
distributed network to achieve an effective and distributed optimization
(Zhang et al., 2022). 1.6. Objective and novelty of the work
Overall, dynamic reconfiguration techniques are stable and capable
of identifying the optimum solution. However, these methods necessi­ The objective of proposed work is to modify the Sudoku configura­
tate a complicated hardware arrangement, which results in an excessive tion so that the shadow is dispersed on the whole photovoltaic array
number of sensors and switches and thus raises the overall system cost. with higher homogeneity. The proposed technique is compared to the
Furthermore, most optimization algorithms exhibit unpredictable Total-Cross-Tied and Sudoku photovoltaic array configurations. For this
behaviour under variable shading conditions. purpose, a 4 × 4 and 9 × 9 PV array and various partial shading sce­
narios are used for the tests. In addition, this paper also implements an
1.4.3. Static array reconfiguration approaches efficient power extraction FES machine learning model based on AI that
Static reconfiguration approaches based on shade dispersion tech­ can assist in predicting the optimum reconfiguration. To evaluate the
niques have been reported in the literature to overcome shortcomings effectiveness of the FES model, a comparison is to be done between
with dynamic reconfiguration approaches. Shade dispersion is a method predicted results from FES model and simulation results of 4 × 4, 5 × 4,
for dispersing the effects of shade across the array to enhance output 6 × 4, 6 × 6 and 9 × 9 PV array from previous research. Following is a
power. One most common Sudoku puzzle-based reconfiguration was summary of the primary novelties of this work as compared to previous
established on a 9 × 9 PV array and found superior in comparison to TCT research on PV array reconfiguration.
(Krishna and Moger, 2019). However, when the Sudoku pattern varies,
the power output changes as well and thus there is a strong probability • The implemented model of FIS system is simple, inexpensive, and
that the Sudoku configuration will produce less output than a standard accurate fuzzy model which can replace conventional prediction
TCT configuration and it is limited to small rated SPV plants. Another methods which depend on complex technological studies. To the best
form of puzzle-based reconfiguration approach reported in the literature of our knowledge, this work is the first to address using an AI-based
is, a novel Zig-Zag technique for TCT configuration (4 × 3), magic FES approach to forecast configurations under different shading
square technique (4 × 4), new enhanced magic square technique to conditions to achieve the most suitable and optimal solution. A wide
enhance the power output of the PV array (Vijayalekshmy et al., 2016; range of input variables and shading conditions can be used to

3
M. Sharma et al. Engineering Applications of Artificial Intelligence 129 (2024) 107516

Fig. 1. Implementation of artificial intelligence techniques in various research areas of SPV system.

construct and apply to proposed machine learning model. It is are applied on a 9 × 9 PV array employing the three configuration
therefore not size-restricted and may be used for both small and large methods TCT, Sudoku, and the proposed M-Sudoku. This work done uses
SPV fields. It can be utilized as a dynamic predictive model to choose the single diode solar cell model used in reference (Kumar and Kumar,
from available reconfigurations techniques of PV array in real time. 2017; Sharma et al., 2023; Ngoc et al., 2019), as this single diode model
• The proposed M-Sudoku reconfiguration, very simple and clear is the most widely used and provides a good combination of simple
which can effectively spread the shadow throughout the array rather structure and accuracy. In addition, it performs better in PSCs and
than using sophisticated mathematical formulations and complex mismatch effects (Ngoc et al., 2019; Sharma et al., 2022; Pareek and
algorithms. The proposed reconfiguration method is economical Dahiya, 2015).
because it works on static reconfiguration techniques and do not
require complex hardware and provides a feasible solution for large 2.1. Implementation of 4 × 4 and 9 × 9 PV array of TCT, Sudoku and M-
PV array and includes a realistic implementation strategy. Sudoku configuration

The rest of the paper is organized as follows: The methodology to TCT links PV arrays in series to form a string, then connects them in
implement the proposed M- Sudoku reconfiguration on 4 × 4 and 9 × 9 parallel to produce ties (rows). The general arrangement of a TCT
PV array is explained in section 2. In this section, description of various configuration for a 4 × 4 and 9 × 9 PV array and electrical connection of
partial shading cases with the results obtained from TCT, Sudoku, and 9 × 9 TCT-PV array is shown in Fig. 2 (a), (d) and (g) respectively.
the proposed reconfiguration are explained. In Section 3, the design of Sudoku is a number placement matrix based on logic in which horizontal
implemented AI-based machine learning (FES) model, and performance row can only contain numbers from 1 to 4 and each number in the 4 × 4
measurement of given model by comparing its results from simulation block can be used only once in case of 4 × 4 array size. When array size is
results is explained. The analysis and discussion about the various as­ 9 × 9, every row, column, and matrix contain the numbers 1 to 9
pects of the proposed methods is concluded and discussed in Section 4. without any repetitions, and there are nine 3 × 3 matrices in total. The
Sudoku-based configuration for a 4 × 4 and 9 × 9 PV array and electrical
2. Methodology connection of 9 × 9 Sudoku PV array depicted in Fig. 2 (b), (e) and (h)
respectively. The proposed M-Sudoku reconfiguration for a 4 × 4 and 9
A simulation system developed in the MATLAB/Simulink environ­ × 9 PV array and electrical connection of 9 × 9 M-Sudoku PV array is
ment is used to examine the performance of three configurations in the shown in Fig. 2 (c), (f) and (i) respectively and explained in next section.
given work. The ten most common shading patterns are applied on a 4 ×
4 PV array employing the three configuration methods TCT, Sudoku,
and the proposed M-Sudoku. Further, three well known shading patterns

4
M. Sharma et al. Engineering Applications of Artificial Intelligence 129 (2024) 107516

Fig. 2. Connection matrix used for implementation of (a) TCT (4 × 4), (b) Sudoku (4 × 4), (c) M-Sudoku configuration (4 × 4), (d) TCT (9 × 9), (e) Sudoku (9 × 9),
(f) M-Sudoku configuration (9 × 9), array structure and electrical connection of (g) TCT (9 × 9), (h) Sudoku (9 × 9), (i) M-Sudoku (9 × 9).

2.2. Rules used for development of proposed M-Sudoku puzzle-based 2.3. Results obtained from TCT, Sudoku and proposed M-Sudoku
configuration configuration on 4 × 4 PV array

The proposed M-Sudoku reconfiguration is a modified version of The maximum power (PM) generated by TCT, Sudoku, proposed M-
conventional Sudoku. The number placement in the proposed M-Sudoku Sudoku configuration on a 4 × 4 PV array under uniform insolation is
based reconfiguration is modified in such a way that the shadow is 3221 Watt, at current (IM) 32.21 A and voltage (VM) 100 V, all three
dispersed more uniformly over the entire array when compared to configurations deliver the same amount of power under Standard test
conventional Sudoku under partial shading. The following rules are condition (STC). The results of given work done are concluded based on
proposed to configure 4 × 4 and 9 × 9 sized M-Sudoku configuration: Performance Enhancement Ratio (PER) which is described by below
equations:
1. Diagonal summation: The objective of both diagonals is to have Performance Enhancement Ratio (PERTCT ) in %: It is the power
nonrepeating column numbers as shown in Fig. 3 (a) 4 × 4 PV array enhancement of proposed reconfiguration as compared to TCT config­
(b) 9 × 9 PV array. uration and is defined as:
2. Subarray: The objective of all four subarrays sized 2 × 2 is to have
PPro − PTCT
nonrepeating column numbers from 1 to 4 in 4 × 4 matrix, and nine PERTCT = ∗100 (1)
PPro
subarrays sized 3 × 3 is to have nonrepeating row numbers from 1 to
9 in 9 × 9 matrix as presented in Fig. 4(a) and (b) respectively. Performance Enhancement Ratio (PERSudoku ) in %: It is the power
enhancement of proposed reconfiguration as compared to Sudoku
Process flow of implementation of proposed reconfiguration (M- configuration and is defined as:
Sudoku) is shown in Fig. 5. An example of random shading and its PPro − PSuDoKu
implementation on TCT, Sudoku, and M-Sudoku reconfiguration are PERSudoku = (2)
PPro
shown in Fig. 6. The proposed M-Sudoku reconfiguration approach
∗100

follows the above-mentioned reconfiguration rules for shade dispersion. Where:

The proposed M-Sudoku design’s PV arrays is rearranged with the aim to
distribute the shade effect more evenly around the PV array.

5
M. Sharma et al. Engineering Applications of Artificial Intelligence 129 (2024) 107516

PTCT = Maximum power obtained from TCT configuration

PPro = Maximum power obtained from proposed M-Sudoku
configuration
PSuDoKu = Maximum power obtained from Sudoku configuration

2.4. Partial shading cases used in current work on a 4 × 4 PV array

The proposed M-Sudoku reconfiguration method has been evaluated

on a 4 × 4 PV array with ten partial shading patterns which are most
relevant and probable: Short narrow (SN), long narrow (LN), long wide
(LW), single row (SR), oblique-1 (OB-1), oblique-2 (OB-2), oblique-3
(OB-3), left bottom corner (LBC), right top corner (RTC), and diagonal
(DN), as shown in Fig. 7.

2.5. Case studies on 4 × 4 PV array

2.5.1. Case 1: Short Narrow (SN) shadow

The PV array is exposed to 3 different (200 w/m2, 400 w/m2, 1000
w/m2) shade types in SN shadow as shown in Fig. 7. It implies that with
proposed PV array reconfiguration, the locations of PV modules in each
column are uniformly dispersed into different rows. Theoretical currents
of all the rows currents for all three configurations can be calculated as
follows:
Row current calculations:
The current produced by whole PV arrays is calculated as IM. Row
current calculation for TCT configuration for shading Case-1 i.e., short
narrow, are as follows:
IR1 = 2 ∗ 0.2∗IM +2∗IM= 2.4IM

Fig. 3. Diagonal summation rule for (a) 4 × 4 PV array (b) 9 × 9 PV array of M- IR2 = 2 ∗ 0.4∗IM +2∗IM = 2.8IM
Sudoku configuration.
IR3 = IR4 = 4IM (3)

Fig. 4. Sub Array summation rule for (a) 4 × 4 PV array (b) 9 × 9 PV array of Modified-Sudoku configuration.

6
M. Sharma et al. Engineering Applications of Artificial Intelligence 129 (2024) 107516

Fig. 5. Process flow of implementation of proposed reconfiguration (M-Sudoku).

Fig. 6. Illustration of shading dispersion of TCT, Sudoku, and M-Sudoku reconfiguration.

Row current calculation for Sudoku configuration: IR4 = 4∗IM = 4IM (4)
IR1 = 1 ∗ 0.2∗IM +1 ∗ 0.4∗IM +2∗IM = 2.6IM The generated current in each string is different as shown in the
above calculations, and the PV curves have multiple peaks. The voltage
IR2 = 1 ∗ 0.4∗IM +3∗IM = 3.4IM
across the array is Varray = 4 × Vm is the voltage across the array, and the
power produced in array is expressed as:
IR3 = 1 ∗ 0.2∗IM +3∗IM = 3.2IM

7
M. Sharma et al. Engineering Applications of Artificial Intelligence 129 (2024) 107516

Fig. 7. Partial shading application example used in current study for 4 × 4 PV array.

Fig. 8. I–V characteristics obtained by TCT, Sudoku, and M-Sudoku reconfiguration under ten shading application examples used in current study for 4 × 4 PV array.

Parray = Varray ×4IM (5) 2629W, respectively. As per the results, M-Sudoku configuration has a
maximum PM of 2629 W with a minimum mismatch loss (ML) of 592 W,
Row current calculation for Proposed M-Sudoku reconfiguration:
and Power loss (Ploss) 18%. Figs. 8 and 9 illustrate, respectively, the
IR1 = 1 ∗ 0.2∗IM +3∗IM = 3.2IM current-voltage & power-voltage characteristics for all ten shading
cases.
IR2 = 1 ∗ 0.4∗IM +3∗IM = 3.4IM The PERTCT and PERSudoku for the SN shadow pattern can be calcu­
lated as:
IR3 = 1 ∗ 0.4∗IM +3∗IM = 3.4IM
2629 − 2134
PERTCT = ∗100 = 18% (7)
IR4 = 1 ∗ 0.2∗IM +3∗IM = 3.2IM (6) 2629

The row currents of all other nine shading cases can be calculated 2629 − 2308
PERSudoku = (8)
similarly. 2629 ∗100=12%

From the results obtained it is observed that the PM of the TCT, The PERTCT and PERSudoku of all other nine shading conditions are
Sudoku, and the M-Sudoku configurations, are 2134 W, 2308 W, and

8
M. Sharma et al. Engineering Applications of Artificial Intelligence 129 (2024) 107516

Fig. 9. P–V characteristics obtained by TCT, Sudoku, and M-Sudoku reconfiguration under ten shading application examples used in current study for 4 × 4 PV array.

Table 2
Summarized table for electrical parameters (PM, VM, IM) obtained by TCT, Sudoku, and M-Sudoku reconfiguration under ten shading application examples used in
current study for 4 × 4 PV array.
Shading Case TCT Sudoku M-Sudoku PERTCT (%) PERSUDOKU (%) Best Configuration

PM (W) VM (V) IM (A) PM (W) VM (V) IM (A) PM (W) VM (V) IM (A)

WS 3221 110 29.2 3221 110 29.2 3221 110 29.2 – ALL
SN 2134 110 19.4 2308 110 20.9 2629 110 23.9 18 12 M-Sudoku
LN 2077 110 18.9 2149 110 19.5 2149 110 19.5 3.3 0 Sudoku/M-Sudoku
LW 1657 110 15.1 1410 110 12.8 1745 110 15.9 5 19 M-Sudoku
SR 2409 80 30.1 2548 110 23.2 2548 110 23.2 5.4 0 Sudoku/M-Sudoku
OB-1 2056 80 25.7 2138 80 26.7 2639 110 24 22 18 M-Sudoku
OB-2 2138 110 19.4 2626 110 23.8 2640 110 24 19 0.5 M-Sudoku
OB-3 1552 80 19.4 1552 80 19.4 2047 80 25.6 24 24 M-Sudoku
LBC 2090 110 19.0 2133 110 19.4 2548 110 23.2 17 16 M-Sudoku
RTC 2090 110 19.0 2133 110 19.4 2548 110 23.2 17 16 M-Sudoku
DN 2074 110 18.9 1557 50 31.2 2125 110 19.3 2.4 26 M-Sudoku

Fig. 10. Partial shading application example used in current study for 9 × 9 PV array.

9
M. Sharma et al. Engineering Applications of Artificial Intelligence 129 (2024) 107516

Fig. 11. Shade dispersion obtained by Sudoku and M-sudoku reconfiguration on a 9 × 9 PV array under short narrow shading case.

calculated similarly. (200 w/m2, 1000 w/m2) groups as shown in Fig. 7. The PM of the TCT,
Sudoku and M-Sudoku configurations are 2409 W, 2548 W, and 2548 W,
2.5.2. Case 2: Long Narrow (LN) shadow respectively. According to the observations, the Sudoku, and M-Sudoku
In the LN shadow, PV array is subjected to four separate (200 w/m2, configurations both have achieved a maximum PM of 2548 W with a
400 w/m2, 600 w/m2, 1000 w/m2) shade groups as shown in Fig. 7. The minimum ML of 673 W and Ploss 20%, PER w.r.t. to TCT is 5.4 %.
PM of the TCT, Sudoku, and M-Sudoku configurations are 2077 W, 2149
W, and 2149 W, respectively. According to the observations, the Sudoku 2.5.5. Case 5: Oblique-1 (OB-1) shadow
and M-Sudoku configurations have both achieved a maximum PM of The PV array is subjected to two separate (200 w/m2, 1000 w/m2)
2149 W with a minimum ML of 1072 W and a power loss (Ploss) of 33%. shade groups in the OB-1 shadow as shown in Fig. 7. The PM of the TCT,
The PER of M-Sudoku and Sudoku configuration is 3.3% with respect to Sudoku, and M-Sudoku configurations are observed as 2056 W, 2138 W,
TCT. and 2639 W, respectively. According to the observations, the M-Sudoku
configurations have achieved a maximum PM of 2639 W with a mini­
2.5.3. Case 3: Long Wide (LW) shadow mum ML of 582, Ploss 21%, PER of 22 % and 18 % w.r.t. TCT and Sudoku
There are four separate (200 w/m2, 400 w/m2, 600 w/m2, 1000 w/ respectively.
2
m ) shade groups applied to the PV array in the LN shadow as shown in
Fig. 7. The PM of the TCT, Sudoku, and M-Sudoku configurations are 2.5.6. Case 6: Oblique-2 (OB-2) shadow
1657 W, 1410 W, and 1745 W, respectively. According to the observa­ The PV array is subjected to two separate (200 w/m2, 1000 w/m2)
tions, the M-Sudoku configurations have achieved a maximum PM of shade groups in the OB-2 shadow as shown in Fig. 7. The PM of the TCT,
1745 W with a minimum ML of 1476 W and Ploss of 45.8%. PER of M- Sudoku, and M-Sudoku configurations are observed as 2138 W, 2626 W,
Sudoku configuration is 5 % w.r.t to TCT and 19% w.r.t. Sudoku. and 2640 W, respectively. According to the observations, the M-Sudoku
configurations have achieved a maximum PM of 2640 W with a mini­
2.5.4. Case 4: Single Row (SR) shadow mum ML of 581 W, Ploss of 18%, and PER 19 % and 0.5 % w.r.t. TCT and
In the SR shadow, the PV array is subjected to two different shade Sudoku respectively.

Table 3
Summarized table for electrical parameters (PM, VM, IM) obtained from TCT, Sudoku, and M-Sudoku reconfiguration under SN, LN and CN shading application ex­
amples used in current study for 9 × 9 PV array.
Shading Case TCT Sudoku M-Sudoku PERTCT (%) PERSUDOKU (%) Best Configuration

PM (W) VM (V) IM (A) PM (W) VM (V) IM (A) PM (W) VM (V) IM (A)

WS 16396 240 68.3 16396 240 68.3 16396 240 68.3 – – ALL
SN 13835 250 55.3 14661 240 61.1 14919 240 62.2 7.2 1.7 M-Sudoku
LN 13500 250 54 14331 240 59.7 14359 250 57.4 6 0.1 M-Sudoku
CN 13593 250 18.9 14430 240 31.2 14543 250 19.3 6.5 0.7 M-Sudoku

10
M. Sharma et al. Engineering Applications of Artificial Intelligence 129 (2024) 107516

Fig. 12. Shade dispersion obtained by Sudoku and M-sudoku reconfiguration on a 9 × 9 PV array under long narrow shading case.

Fig. 13. Shade dispersion obtained by Sudoku and M-sudoku reconfiguration on a 9 × 9 PV array under centre shading case.

2.5.7. Case 7: Oblique-3 (OB-3) shadow Sudoku, and M-Sudoku configurations are 1552 W, 1552 W, and 2047
Two distinct shade (200 w/m2, 1000 w/m2) groups are applied to the W, respectively. According to the results obtained, the M-Sudoku con­
PV array in the OB-3 shadow as shown in Fig. 7. The PM of the TCT, figurations achieved a maximum PM of 2047 W with a minimum ML of

11
M. Sharma et al. Engineering Applications of Artificial Intelligence 129 (2024) 107516

Fig. 14. I–V characteristics obtained by TCT, Sudoku, and M-Sudoku reconfiguration under SN, LN and CN shading application examples used in current study for 9
× 9 PV array.

Fig. 15. P–V characteristics obtained by TCT, Sudoku, and M-Sudoku reconfiguration under SN, LN and CN shading application examples used in current study for 9
× 9 PV array.

1174 W, Ploss 36.4 %, and PER 24% w.r.t. both TCT and Sudoku has a maximum PM of 2548 W with a minimum ML 673 W, Ploss 20.8%,
configurations. and a PER of 17 % and 16% w.r.t. TCT and Sudoku configurations
respectively.
2.5.8. Case 8: Left Bottom Corner (LBC) shadow
The PV array is subjected to two distinct shade groups (200 w/m2, 2.5.10. Case 10: Diagonal (DN) shadow
1000 w/m2) in the LBC shadow as shown in Fig. 7. The PM of the TCT, The PV array is subjected to two separate shade (200 w/m2, 1000 w/
Sudoku, and M-Sudoku configurations are 2090 W, 2133 W, and 2548 m2) groups in the DN shadow as shown in Fig. 7. The PM of the TCT,
W, respectively. According to the results obtained, the M-Sudoku con­ Sudoku, and M-Sudoku configurations are 2074 W, 1557 W, and 2125
figurations achieved a maximum PM of 2548 W with a minimum ML 673 W, respectively. The M-Sudoku configurations achieved a maximum PM
W, Ploss 20.8%, and a PER of 17 % and 16% w.r.t. TCT and Sudoku of 2125 W with minimum ML of 1096 W, Ploss 34%, and a PER 2.4 % and
configurations respectively. 26% w.r.t. TCT and Sudoku respectively.
The electrical parameters PM, VM and IM obtained by TCT, Sudoku,
2.5.9. Case 9: Right Top Corner (RTC) shadow and M-Sudoku for all ten shading cases as discussed above are
Two distinct shade (200 w/m2, 1000 w/m2) groups in the RTC mentioned in Table 2.
shadow are used in PV array as shown in Fig. 7. The PM of the TCT,
Sudoku, and M-Sudoku configurations are 2090 W, 2133 W, and 2548
W, respectively. According to the results, the M-Sudoku configurations

12
M. Sharma et al. Engineering Applications of Artificial Intelligence 129 (2024) 107516

Fig. 16. Generalized Mamdani Fuzzy Expert System layout and processing.

2.6. Results obtained from TCT, Sudoku and proposed M-Sudoku

configuration on 9 × 9 PV array

Under STC and uniform irradiance, the maximum power (PM) pro­
duced by TCT, Sudoku, and the M-Sudoku reconfiguration on a 9 × 9 PV
array is 16397 Watts, at current (IM) 68.3 A and voltage (VM) 240 V.

2.7. Partial shading cases used in current work

A 9 × 9 PV array is subjected to the proposed reconfiguration

approach with three different shading patterns: short narrow (SN), long
narrow (LN), and centre (CN) as shown in Fig. 10. Additionally, TCT and
Sudoku methods are investigated and compared with the proposed Fig. 18. A set of 6 membership functions defined for input variable “Position
reconfiguration method. of shading”.

2.8. Case studies on 9 × 9 PV array Row current calculation for TCT configuration:
IR1 = IR2 = IR3 = IR4 = IR5 = 9IM
2.8.1. Case 1: Short Narrow (SN) shadow pattern
The PV array is subjected to three separate shade groups (400 w/m2, IR6 = IR7 = 5IM +2 × 0.6IM = 2 × 0.4IM = 7IM
600 w/m2, 1000 w/m2) in the SN shadow pattern. In this type of
shading, only four columns and four rows are exposed to shading, and it IR8 = IR9 = 5IM +4 × 0.6IM = 7.4IM (9)
is referred as a SN shadow pattern. According to the proposed reconfi­
Row current calculation for Sudoku configuration:
guration of the PV array, as illustrated in the Fig. 11, the PV module
positions in each column are uniformly distributed into various rows in IR1 = IR6 = IR7 = 8IM +0.6IM = 8.6IM
order to reduce the current differences between shaded and unshaded
rows of PV array. The following calculations can be used to determine IR2 = IR3 = IR4 = 7IM +2 × 0.6IM = 8.2IM
the theoretical currents for all the rows currents:

Fig. 17. A set of 16 membership functions defined for input variable “Type of shading”.

13
M. Sharma et al. Engineering Applications of Artificial Intelligence 129 (2024) 107516

Watts as mentioned in the Table 3. I–V and P–V characteristics obtained

by TCT, Sudoku, and M-Sudoku reconfiguration under SN, LN and CN
shading application examples used in current study for 9 × 9 PV array is
shown in Figs. 14 and 15 respectively.

2.8.2. Case 2: Long Narrow (LN) shadow pattern

The PV array is subjected to four separate shade (300 w/m2, 400 w/
m2, 700 w/m2, 1000 w/m2) groups in the LN shadow pattern. In this
type of shading, only four columns and seven rows are exposed to
shading, and it is referred to as LN shadow pattern. The shade dispersion
from proposed M-Sudoku reconfiguration, and Sudoku configurations
are shown in Fig. 12. The PM achieved by TCT is 13500 Watts, Sudoku is
14331 Watts, and the proposed M-Sudoku reconfiguration is 14359
Fig. 19. A set of 5 membership functions defined for input variable Watts. The proposed reconfiguration is producing more power than the
“Array size”. TCT and Sudoku configurations under the LN shadow pattern. It’s also
worth noting that the recommended configuration has a 7.1% PERTCT ..

2.8.3. Case 3: Centre (CN) shadow pattern

The PV array is subjected to five different shade (200 w/m2, 400 w/
m2, 600 w/m2, 800 w/m2, 1000 w/m2) groups in the CN shadow
pattern. In this type of shading, four rows and four columns with panels
in the centre are exposed to shading, and it is referred as a centre shadow
pattern. The shading dispersion from proposed M-Sudoku reconfigura­
tion, and Sudoku reconfigurations are shown in Fig. 13. The PM achieved
by TCT is 13,593 Watts, Sudoku is 14,430 Watts, and the proposed M-
Sudoku method is 14,543 Watts. It may be seen that the proposed
reconfiguration is producing better power compared to TCT and Sudoku
configurations under the CN shadow pattern.
Overall, this work done investigates results of 39 simulations (3
configurations, 10 shading patterns for 4 × 4 PV array as mentioned in
Fig. 20. A set of 10 membership functions defined for output variable “Best Table 2 and 3 configurations, 3 shading patterns for 9 × 9 PV array) as
configuration”. mentioned in Table 3. Under a major shading scenario, when compared
to TCT and Sudoku configurations, the proposed M-Sudoku reconfigu­
ration enhances PM and reduces ML thus, may be considered as an ideal
IR5 = 6IM +2 × 0.6IM +0.4IM = 7.6IM
choice for PV systems during panel installation of solar industries.
IR8 = 7IM +0.6IM +0.4IM = 8IM
3. Implementation of artificial intelligence-based machine
IR9 = 7IM +2 × 0.4IM = 7.8IM (10) learning model

Row current calculation for Proposed M-Sudoku for reconfiguration: FES is the most common AI-based deep learning model because it is a
IR1 = IR3 = IR4 = IR7 = 8IM +1 × 0.6IM = 8.6IM robust and flexible system with a simple structure that is easy to build.
The Fuzzy logic systems have similar logic like human, thus can solve
IR2 = 7IM +2 × 0.6IM = 8.2IM complex problems. It can also be carried out using hardware, software,
or a combination of both. Considering the numerous benefits associated
IR5 = 6IM +0.6IM +2 × 0.4IM = 7.4IM with FES, it is employed in present work.

IR6 = IR9 = 7IM +0.6IM +0.4IM = 8IM 3.1. Fuzzy Expert Systems for prediction of best PV configuration

IR8 = 6IM +3 × 0.6IM = 7.8IM (11) FES is a modelling structures with precisely defined functional input
and output blocks, as well as a processing engine that does all the
The row currents of long narrow and centre shading cases can be
computation at the language level which is normally used by human
calculated similarly.
being. Mamdani fuzzy inference is used to develop FES system in this
The PERTCT and PERSudoku for the SN shadow pattern can be calcu­
work. Fuzzification interface, knowledge base, decision-making unit,
lated as:
and defuzzification interface are the four general components of Mam­
PERTCT =
14919 − 13835
∗100 = 7.2% (12) dani fuzzy inference system, as depicted in Fig. 16. An implemented FES
14919 includes use of data from current work done described in previous sec­
tions of this paper for 4 × 4 PV array. In addition, this implemented
14919 − 14661
PERSudoku = (13) model is also tested on 5 × 4 (Pachauri et al., 2020), 6 × 4 (Shams
14919 ∗100=1.7 %
El-et al., 2013), 6 × 6 (Pachauri et al., 2020) and 9 × 9 (Sharma et al.,
The PERTCT and PERSudoku of long narrow and centre shading cases are 2023) which uses a simulation result for creating knowledge base of
calculated similarly. proposed model. In this study, best configuration is an output, while
It is possible to conclude from the above row current calculations for type of shading and position of shading and size of array are inputs. The
TCT, Sudoku, and the proposed M-Sudoku reconfiguration that the system then evaluates the rules to generate the output based on the
suggested reconfiguration has a smaller current difference than TCT. inputs.
Furthermore, TCT achieves a PM of 13835 Watts, while Sudoku achieves
a PM of 14661 Watts, which is less than the proposed method’s 14919

14
M. Sharma et al. Engineering Applications of Artificial Intelligence 129 (2024) 107516

Table 4
Validation of implemented FES machine learning model.
Array Case Shading Pattern Best Configuration according to Simulation Best Configuration according to FIS Results of FES matches with Simulation
Size results results results

4£4 Case: 1 Without Shading ALL 9 (ALL) yes

Case: 2 Short Narrow M-Sudoku 3 (M-Sudoku) yes
Case: 3 Long Narrow Sudoku/M-Sudoku 4 (Sudoku/M-Sudoku) yes
Case: 4 Long Wide M-Sudoku 3 (M-Sudoku) yes
Case: 5 Single Row Sudoku/M-Sudoku 4 (Sudoku/M-Sudoku) yes
Case: 6 Oblique-1 M-Sudoku 3 (M-Sudoku) yes
Case: 7 Oblique − 2 M-Sudoku 3 (M-Sudoku) yes
Case: 8 Oblique − 3 M-Sudoku 3 (M-Sudoku) yes
Case: 9 Left Bottom M-Sudoku 3 (M-Sudoku) yes
Corner
Case: Right Top Corner M-Sudoku 3 (M-Sudoku) yes
10
Case: Diagonal M-Sudoku 3 (M-Sudoku) yes
11
6£4 Case: Without Shading ALL 9 (ALL) yes
12
Case: Short Narrow TCT 1 (TCT) yes
13
Case: Long Wide TCT 1 (TCT) yes
14
5£4 Case: Without Shading ALL 9 (ALL) yes
15
Case: Center HC 6 (HC) yes
16
Case: Top right corner TCT 1 (TCT) yes
17
6£6 Case: Without Shading ALL 9 (ALL) yes
18
Case: Short Narrow TCT 1 (TCT) yes
19
Case: Long Narrow TCT 1 (TCT) yes
20
Case: Long Wide TCT 1 (TCT) yes
21
Case: Short Wide TCT 1 (TCT) yes
22
Case: Diagonal HC 6 (HC) yes
23
Case: Outer TCT 1 (TCT) yes
24
Case: Multi-storied TCT 1 (TCT) yes
25
Case: Single row ALL 9 (ALL) yes
26
Case: Half Array ALL 9 (ALL) yes
27
Case: L-Shape HC 6 (HC) yes
28
9£9 Case: Without Shading ALL 9 (ALL) yes
29
Case: Short Narrow Robust-TCT 8 (Robust-TCT) yes
30
Case: Short Wide Robust-TCT/Sudoku 10 (Robust-TCT/Sudoku) yes
31
Case: Long Narrow Robust-TCT 8 (Robust-TCT) yes
32
Case: Center Robust-TCT 8 (Robust-TCT) yes
33

3.2. Developing the proposed Fuzzy Expert Systems for prediction of best used triangular membership functions (MFs) due to their simplicity.
PV configuration These MFs also incorporate simple formulas and effective computational
methods. Additionally, they include a segment of straight lines, making
The first step of constructing the proposed Mamdani fuzzy model is it simple to implement them in Fuzzy Systems.
to define the input and output variables. Commonly, fuzzy variables are
used to define input and output variable while creating a fuzzy model for
a specific system. Each crisp variable’s value is fuzzified, producing the 3.3. Fuzzification of input and output variable
fuzzy values of the fuzzy variable. These fuzzy values typically consist of
a pair of components with the first component being a linguistic value In proposed FIS system, the three variables: size of array, type of
and the second component being the degree of membership in each shading and position of shading are used as input. The proposed FES
linguistic value (which is typically between 0 and 1). This work done has system has been designed to function using multiple PV array sizes, so
that it may be used to optimize configuration selection among a wide

15
M. Sharma et al. Engineering Applications of Artificial Intelligence 129 (2024) 107516

3.4. Knowledge-based fuzzy rules of the model

The rule base of a FES is a key component, and effective fuzzy rules
are primarily determines the accuracy. The rules are created and
modified using the Rule Editor. After fuzzification, the rules set converts
fuzzy inputs into fuzzy outputs. Fuzzy inference rules are defined based
on logical implications. In this work, IF-THEN rules are applied to each
fuzzy system through IF < input variable1> AND < input variable 2>
AND < input variable 3> THEN < output variable>.
Number of rules is calculated as follows:
∏n
R = i=1 FSi (14)

where R is the number of rules, n is the number of input variables, and

FSi denotes the number of fuzzy sets assigned to the i’th input variable.
Theoretically the number of the rules directly depends on the number of
input variables and fuzzy sets in the system. However, the formulation of
numerous rules does not appear feasible or reasonable, and it increases
the possibility of errors being generated when interpreting the results.
Additionally, this has a significant computational cost, which limits the
model’s application for systems with fast dynamics. Therfore, this work
created a set of 45 optimal rules to minimize complexity as mentioned
below:

Rule 1 If “Type of shading” is without shading and “Position of shading” is

none and “Array size” is 4 × 4, then “Best Configuration” is All.
Rule 2 If “Type of shading” is short narrow and “Position of shading” is
left, and Array size” is 4 × 4, then “Best Configuration” is M-
Sudoku.
Rule 3 If “Type of shading” is long narrow and “Position of shading” is left
and Array size” is 4 × 4, then “Best Configuration” is Sudoku and
M-Sudoku.
Rule 12 If “Type of shading” is short narrow and “Position of shading” is
left and Array size” is 6 × 4, then “Best Configuration” is TCT.
Rule 13 If “Type of shading” is long wide and “Position of shading” is left
and Array size” is 6 × 4, then “Best Configuration” is TCT.
Rule 23 If “Type of shading” is center and “Position of shading” is center
and Array size” is 5 × 4 and then “Best Configuration” is HC.
Rule 24 If “Type of shading” is top right corner and “Position of shading” is
right and Array size” is 5 × 4, then “Best Configuration” is TCT.
Rule 45 If “Type of shading” is short wide and “Position of shading” is
bottom and Array size” is 9 × 9, then “Best Configuration” is
Robust-TCT/Sudoku.

3.5. Defuzzification method for implemented FES model

The defuzzification method for current work is the Centroid method

which is explained in equation (15). This method provides a crisp value
based on the centre of gravity of the fuzzy set. The total area of the
membership function distribution used to represent the combined con­
Fig. 21. Rule viewer that displays the output prediction of the FES model for trol action and divided into several sub-areas. The area and centroid of
(a) Case 4, (b) Case 16, (c) Case 25, (d) Case 30. each sub-area is calculated and then the summation of all these sub-
areas is taken to find the defuzzified value for a discrete fuzzy set.
range of applications and is not limited to a single array size. The input ∑
n
variable size of array in the current work is defined for 4 × 4, 5 × 4, 6 × xi .μ(xi )
4, 6 × 6 and 9 × 9 PV array as shown in Fig. 19. For input variable type x∗ = i=1
∑
n (15)
of shading, 16 fuzzy sets are created as illustrated in Fig. 17. Six fuzzy μ(xi )
sets are created from the input variable position of shading as shown in
i=1

Fig. 18. Proposed FIS model objective is to predict the best configuration The defuzzified value denoted a sx∗ , xi indicates the sample element,
for a specific shading type and position. The best configuration, an μ(xi ) is the membership function, and n represents the number of ele­
output variable with ten fuzzy sets is mentioned in Fig. 20. Fuzzification ments in the sample. After the defuzzification, result obtained of the FES
and degree of membership and range of input and output variable used is a set of fuzzy values that are created by combining the results of all
in FES model is shown in Figs. 17–20. membership functions used by the inference engine.

16
M. Sharma et al. Engineering Applications of Artificial Intelligence 129 (2024) 107516

Fig. 22. Flow chart of Implementation AI-based FES system for prediction of best configuration.

3.6. Results obtained from FES forecast(i) = refers to predicted results of best configuration from
FES model implemented in current study.
All 33 shading cases mentioned in Table 4 are examined with the n = number of sample or test data used for prediction via FES model.
implemented fuzzy model. However, only few cases (Case 4, Case 16, In this case 33 samples are used as shown in Table 4.
Case 25, Case 30) results as a rule viewer are presented in the Fig. 21 due actual(i) − forecast(i) = difference between (the actual results of best
to space limitations. To evaluate the accuracy of FES, a comparison is configuration obtained from simulation results of previous research
made between the output produced by the FES and the simulation re­ used to create knowledge base) and (predicted results of best
sults obtained from 4 × 4, 5 × 4, 6 × 4, and 6 × 6 PV arrays as shown in configuration from FES model implemented in current study).
Table 4 and then calculated Mean Absolute Percentage Error. 33 ( )
∑
Error Calculation. MAPE =
33 − 33
∗100 = 0(NIL) (17)
Mean Absolute Percentage Error (MAPE): It is considering one of the i=1
33
most popular error measurements which evaluate the accuracy of pre­
Results demonstrated that the implemented FES model generated
diction. It measures the average deviation of predictions from actual
accurate results and 0% MAPE (actual(i) = forecast(i) in all 33 shading
values, which is expressed as a percentage.
conditions) in the best suitable configuration prediction used for model
n ( )
testing. A 0% MAPE indicates that the forecasts perfectly match the
1∑ actual(i) − forecast(i)
MAPE = ∗100 (16) actual values, and it further implies that the forecasting model has
n i=1 actual(i)
achieved a high level of accuracy which can be considered as a signifi­
Where: cant achievement. Results of implemented FES model demonstrates that
it can be considered as potential model which provide reliable and ac­
actual(i) = refers to actual results of best configuration obtained curate predictions. The achievement of a 0% MAPE in this work in­
from simulation results of previous research used to create knowl­ dicates the potential of AI-based forecasting models for achieving high
edge base. levels of accuracy. However, it is crucial to recognize that real-world

17
M. Sharma et al. Engineering Applications of Artificial Intelligence 129 (2024) 107516

identified as AI based controllers which can adapt to dynamic weather

conditions and optimize energy extraction from PV panels in real-time.
AI may also facilitate seamless integration of PV systems into smart
grids, enabling real-time demand response and grid optimization. Re­
searchers may also conduct field experiments and simulations to eval­
uate the performance of AI algorithms under varying weather conditions
and grid dynamics. Future research may also focus on developing
advanced AI algorithms, optimizing hardware architectures, and vali­
dating their performance in real-world conditions. These advancements
will establish the future energy system by intelligent PV systems that
Fig. 23. FES model for dynamic reconfigurable PV arrays. contribute to a sustainable and renewable energy landscape. The
possible approach for implementation of the model under variable
systems are complex and dynamic, continuous monitoring, evaluation, conditions which will enable optimal application is illustrated in Fig. 23.
and refinement of the forecasting model are essential to ensure its
continued effectiveness in a changing environment. Validation of CRediT authorship contribution statement
implemented FES results with simulation results of current work done is
shown in Table 4. The implemented FES system model offers several Mona Sharma: Conceptualization, Methodology, Software, Valida­
benefits, such as simplicity, interpretability, affordability, and fast tion, Formal analysis, Investigation, Writing – original draft, Writing –
learning. Its computational base is built on a simple set of fundamental review & editing, Visualization. Smita Pareek: Conceptualization,
rules. The FES model for prediction of a suitable configuration can Methodology, Formal analysis, Writing – review & editing, Supervision,
handle a complex technical concept which is difficult for a non-technical Project administration. Kulwant Singh: Conceptualization, Methodol­
person. Once the model is created, user interface is relatively simple thus ogy, Formal analysis, Writing – review & editing, Supervision, Project
non-technical and less skilled person can also use it as a replacement of administration.
conventional complex techniques for installation of solar panels. Flow
chart of Implementation AI-based FES system for prediction of best
configuration is shown in Fig. 22. Declaration of competing interest

4. Conclusion The authors declare that they have no known competing financial
interests or personal relationships that could have appeared to influence
This paper deals with the extraction of maximum power from solar the work reported in this paper.
photovoltaic systems under partial shading conditions by using recon­
figurable photovoltaic systems. The paper first proposes, a modified Data availability
Sudoku (M-Sudoku) reconfiguration based on static techniques. The
implementation of the proposed M-Sudoku on 4 × 4 and 9 × 9 PV array Data will be made available on request.
with the objective to modify the Sudoku configuration so that the
shadow is dispersed on the whole photovoltaic array with higher ho­ Acknowledgement
mogeneity. The proposed technique is compared to the most prevalent
and well-known static configuration that is Total-Cross-Tied and Sudoku The authors acknowledge the computational facilities provided by
photovoltaic array configurations. The results of 4 × 4 PV array Multiscale Simulation Research Center (MSRC) at Manipal University
demonstrate that the proposed M-Sudoku enhances PM in all ten shading Jaipur.
cases under study when compared to the TCT configuration, with a
maximum improved PER 24% in oblique shading. Additionally, the References
proposed reconfiguration has better PM in eight shading cases and equal
PM in two shading cases when compared to the Sudoku configuration, Ahm, D., Hamdy, S., Ton, D., Omar, K., Mossa, Mahmoud, 2020. “Coyote Optimization
Algorithm for Parameters Estimation of Various Models of Solar Cells and PV
with a maximum improved PER of 26% in diagonal shading. In case of 9
Modules”. IEEE Access.
× 9 PV array the proposed reconfiguration improves power and reduced Babu, T.S., Rajasekar, N., Sangeetha, K., 2015. Modified particle swarm optimization
ML under all three shading scenarios when compared to TCT and technique based maximum power point tracking for uniform and under partial
Sudoku configurations with improved PER of 7.2% in case of short shading condition. Appl. Soft Comput. 34, 613–624.
Bayoumi, A.S.A., El-Sehiemy, R.A., Abaza, A., 2022. Effective PV parameter estimation
narrow shading pattern when compared to TCT configuration. algorithm based on marine predators optimizer considering normal and low
Furthermore, the proposed method does not require a complicated al­ radiation operating conditions. Arabian J. Sci. Eng. 47, 3089–3104. https://doi.org/
gorithm, switching matrix, or sensors, making it more cost-effective. In 10.1007/s13369-021-06045-0.
Dalia, Y., Babu, T.S., Karthik, B., Ahmed, O., Ahmed, F., 2020. “Multi-Objective Grey
the second stage of this work, a FES model for predicting the best Wolf Optimizer for Optimal Design of Switching Matrix for Shaded PV Array
configuration for PV system under PSC is developed using AI based Dynamic Reconfiguration”. IEEE Access, 1-1.
machine learning techniques. To evaluate the performance of the FES Das, S., Verma, D., Nema, S., 2017. Shading mitigation techniques: state-of-the-art in
photovoltaic applications. Renew. Sustain. Energy Rev. 78, 369–390.
model, simulations were carried out and a comparison was made be­ Dhanalakshmi, B., Rajasekar, N., 2018a. “Dominance Square-Based Array
tween the predicted results and the results obtained from the simula­ Reconfiguration Scheme for Power Loss Reduction in Solar Photovoltaic (Pv)
tions on 5 × 4, 6 × 4, 6 × 6 and 9 × 9 PV array. Results demonstrated Systems”, vol. 156. Energy conversion and management, pp. 84–102.
Dhanalakshmi, B., Rajasekar, N., 2018b. A novel competence square based pv array
that the implemented FES model generated accurate results and 0% reconfiguration technique for solar pv maximum power extraction. Energy Convers.
MAPE in all 33 cases of simulations for predicting the best suitable Manag. 174, 897–912.
configuration. Future research in this area may focus on developing Fathy, A., 2018. Recent meta-heuristic grasshopper optimization algorithm for optimal
reconfiguration of partially shaded pv array. Sol. Energy 171, 638–651.
intelligent PV systems that may be utilized for control, supervision, di­
Giovanni, S., Petrone, G., Lehman, B., Paja, C.A.R., Zhao, Y., Gutierrez, M.L.O., 2015.
agnostics, and monitoring in PV applications is the deployment of AI Control of photovoltaic arrays: dynamical reconfiguration for fighting mismatched
techniques based on hardware circuits like microcontroller and pro­ conditions and meeting load requests. IEEE Indust. Electronic. Magazine 9 (1),
cessors. The integration of AI techniques with hardware circuits holds 62–76.
Gosumbonggot, J., Fujita, G., 2019. Partial shading detection and global maximum
immense potential for enhancing the efficiency, reliability, and perfor­ power point tracking algorithm for photovoltaic with the variation of irradiation and
mance of SPV systems. Specific avenues for future research may be temperature. Energies 12.

18
M. Sharma et al. Engineering Applications of Artificial Intelligence 129 (2024) 107516

Guillermo, Q., Vera, N., José, J., Francesc, G., Robert, P., 2005. “Irradiance Equalization Shams El-Dein, M.Z., Kazerani, M., Salama, M.M.A., 2013. Optimal photovoltaic array
Method for Output Power Optimization in Plant-Oriented Grid-Connected PV reconfiguration to reduce partial shading losses. IEEE Trans. Sustain. Energy 4 (1),
Generators”. 145–153. https://doi.org/10.1109/TSTE.2012.2208128.
Gupta, A., Gupta, S., Kumar, S., Saxena, R., Kavita, 2022. A comprehensive survey on Sharma, M., Pareek, S., Singh, K., 2020. “Performance analysis of various configurations
role of artificial intelligence in solar energy processes. IEEE 7th Int. Conf. Converg. of solar photovoltaic arrays to mitigate the effects of partial shading conditions. AIP
Tech. (I2CT) 1–6. Conf. Proc. 2294, 040008.
Hyok, J., Jung, D., Kim, J., Won, Chung-Yuen, Oh, Dong-Sung, 2010. Dual mode Sharma, M., Pareek, S., Singh, K., 2022. Comparative review and performance evaluation
switching strategy of flyback inverter for photovoltaic AC modules. IEEE Int. Power of various solar PV system reconfigurations. Asian J. Water Environ. Pollut. 19 (5),
Electronic. Conf. (IPEC) 2924–2929. 31–36.
Jazayeri, M., Jazayeri, K., Uysal, S., 2017. Adaptive photovoltaic array reconfiguration Sharma, M., Pareek, S., Singh, K., 2023. Robust reconfiguration strategies for maximum
based on real cloud patterns to mitigate effects of non-uniform spatial irradiance output power from large-scale photovoltaic arrays under partial shading conditions.
profiles. Sol. Energy 155, 506–516. Physica Scripta J.
Khuran, N., Pareek, S., 2022. Insolation Effect on Solar Photovoltaic Performance Sridharan, M., 2021. Short review on various applications of fuzzy logic-based expert
Parameters. Recent Advances in Industrial Production. Lecture Notes in Mechanical systems in the field of solar energy. Int. J. Ambient Energy 43, 1–45.
Engineering Springer. Verma, D., Pal, D., Bajpai, P., 2019. Comprehensive performance evaluation of various
Kouaissah, N., Hocine, A., 2020. Optimizing sustainable and renewable energy portfolios solar PV system configurations. IET Renew. Power Gener. 13, 1261–1270.
using a fuzzy interval goal programming approach. J. Comput. Indust. Eng. 144. Vijayalekshmy, S., Bindu, G.R., Iyer, S.R., 2016. A novel Zig-Zag scheme for power
Krishna, G.S., Moger, T., 2019. Improved sudoku reconfiguration technique for total- enhancement of partially shaded solar arrays. Sol. Energy 135, 92–102.
cross-tied pv array to enhance maximum power under partial shading conditions. Villa, L.F.L., Ho, T., Crebier, J.C., Raison, B., 2013. A power electronics equalizer
Renew. Sustain. Energy Rev. 109, 333–348. application for partially shaded photovoltaic modules. IEEE Trans. Ind. Electron. 60
Kumar, M., Kumar, A., 2017. Performance assessment and degradation analysis of solar (3), 1179–1190.
photovoltaic technologies: a review. Renew. Sustain. Energy Rev. 78, 554–587. Winston, P.D., Kumaravel, S., Kumar, B.P., Devakirubakaran, S., 2020. Performance
Machina, S.P.C., Koduru, S.S., Madichetty, S., 2022. Solar energy forecasting using deep improvement of solar PV array topologies during various partial shading conditions.
learning techniques. 2nd Int. Conf. Power Electronic. IoT Applicat. Renew. Energy Sol. Energy 196, 228–242.
Control (PARC) 1–6. Wu, Y., Xu, C., Ke, Y., Tao, Y., Li, X., 2019. Portfolio optimization of renewable energy
Mellit, Adel, Kalogirou, Soteris A., 2008. Artificial intelligence techniques for projects under type-2 fuzzy environment with sustainability perspective. J. Comput.
photovoltaic applications: a review. Prog. Energy Combust. Sci. 574–632. Indust. Eng. 133, 69–82.
Nasiruddin, I., Khatoon, S., Jalil, M.F., Bansal, R., 2019. Shade diffusion of partial shaded Yang, B., Shao, R., Zhang, M., et al., 2021. Socio-inspired democratic political algorithm
pv array by using odd-even structure. Sol. Energy 181, 519–529. for optimal PV array reconfiguration to mitigate partial shading. Sustain. Energy
Ngoc, T.N., Quang, N.N., Duy, L.B., 2019. Reconfiguration of solar panels: mathematical Technol Assess Sep. 48, 101627.
model and analysis. Am. J. Electr. Power Energy Syst. 8 (No. 5), 104–110. Yap, K.Y., Sarimuthu, C.R., Lim, J.M.-Y., 2020. Artificial intelligence based MPPT
Nivedha, S., Vijayalaxmi, M., 2021. Performance analysis of fuzzy based hybrid MPPT techniques for solar power system: a review. J. Modern Power Syst. Clean Energy 8
algorithm for photovoltaic system. Proc. 2021 Int. Conf. Commun. Control, Inform. (6), 1043–1059.
Sci. (ICCISc) 1, 1–4. Yousri, D., Allam, D., Eteiba, Magdy B., 2020a. “Optimal Photovoltaic Array
Pachauri, R.K., et al., 2020. Impact of partial shading on various PV array configurations Reconfiguration for Alleviating the Partial Shading Influence Based on a Modified
and different modeling approaches: a comprehensive review. IEEE Access 8, Harris Hawks Optimizer”, vol. 206. Energy Conversion and Management.
181375–181403. https://doi.org/10.1109/ACCESS.2020.3028473. Yousri, D., Mirjalili, T.S., Rajasekar, N., Abd Elaziz, M., 2020b. “A Novel Objective
Pareek, S., Dahiya, R., 2015. Simulation and performance analysis of individual module Function with Artificial Ecosystem-Based Optimization for Relieving the
to address partial shading parameter variation in large photovoltaic fields. J. Energy Mismatching Power Loss of Large-Scale Photovoltaic Array”, vol. 225. Energy
3, 99–104. Conversion and Management.
Pareek, S., Dahiya, R., 2016. Enhanced power generation of partial shaded photovoltaic Zhang, Xiaoshun, Li, Chuanzhi, Li, Zilin, Yin, Xueqiu, Yang, Bo, Gan, Lingxiao, Tao, Yu,
fields by forecasting the interconnection of modules. Energy 95, 561–572. 2021a. Optimal mileage-based PV array reconfiguration using swarm reinforcement
Rajan, N.A., Shrikant, K.D., Dhanalakshmi, B., Rajasekar, N., 2017. Solar pv array learning. Energy Convers. Manag. 232, 113892 https://doi.org/10.1016/j.
reconfiguration using the concept of standard deviation and genetic algorithm. enconman.2021.113892.
Energy Proc. 117, 1062–1069. Zhang, Xiaoshun, Li, Chuanzhi, Li, Zilin, Yin, Xueqiu, Yang, Bo, Gan, Lingxiao, Tao, Yu,
Reddy, S.S., Yammani, C., 2021. A novel Magic-Square puzzle based one-time PV 2021b. Optimal mileage-based PV array reconfiguration using swarm reinforcement
reconfiguration technique to mitigate mismatch power loss under various partial learning. Energy Convers. Manag. 232, 113892.
shading conditions. Optik 222. Zhang, Xiaoshun, Tao, Yu, Ma, Xiaoguang, Guo, Lexin, 2022. An efficient multi-agent
Richard, H., Thornton, C., Dhople, S., Zhao, Z., Sridhar, N., Freeman, D., 2011. High negotiation algorithm for multi-period photovoltaic array reconfiguration with a
efficiency wide load range buck/boost/bridge photovoltaic microconverter. Twenty- hydrogen energy storage system. Energy Convers. Manag. 256, 115376.
Sixth Annual IEEE Appl. Power Electronic. Conf. Exposition (APEC) 309–313.

19