electronics

Article
Optimizing Large-Scale PV Systems with Machine Learning:
A Neuro-Fuzzy MPPT Control for PSCs with Uncertainties
Asif 1 , Waleed Ahmad 1 , Muhammad Bilal Qureshi 1 , Muhammad Mohsin Khan 2 ,
Muhammad A. B. Fayyaz 3, * and Raheel Nawaz 4

1 Department of Electrical and Computer Engineering, COMSATS University Islamabad, Abbottabad Campus,
Abbottabad 22060, Pakistan; [email protected] (M.B.Q.)
2 Sino-Pak Center for Artificial Intelligence (SPCAI), Pak-Austria Fachhochschule-Institute of Applied Sciences
and Technology (PAF-IAST), Haripur 22620, Pakistan
3 Department of Operations Technology, Events and Technology Management, Manchester Metropolitan
University, Manchester M15 6BH, UK
4 Pro Vice Chancellor (Digital Transformation), Staffordshire University, Stoke-on-Trent ST4 2DE, UK
* Correspondence: [email protected]

Abstract: The article proposes a new approach to maximum power point tracking (MPPT) for
photovoltaic (PV) systems operating under partial shading conditions (PSCs) that improves upon
the limitations of traditional methods in identifying the global maximum power (GMP), resulting
in reduced system efficiency. The proposed approach uses a two-stage MPPT method that employs
machine learning (ML) and terminal sliding mode control (TSMC). In the first stage, a neuro fuzzy
network (NFN) is used to improve the accuracy of the reference voltage generation for MPPT, while in
the second stage, a TSMC is used to track the MPP voltage using a non-inverting DC—DC buck-boost
converter. The proposed method has been validated through numerical simulations and experiments,
demonstrating significant enhancements in MPPT performance even under challenging scenarios. A
comprehensive comparison study was conducted with two traditional MPPT algorithms, PID and
P&O, which demonstrated the superiority of the proposed method in generating higher power and
less control time. The proposed method generates the least power loss in both steady and dynamic
Citation: A.; Ahmad, W.; Qureshi,
states and exhibits an 8.2% higher average power and 60% less control time compared to traditional
M.B.; Khan, M.M.; Fayyaz, M.A.B.; methods, indicating its superior performance. The proposed method was also found to perform well
Nawaz, R. Optimizing Large-Scale under real-world conditions and load variations, resulting in 56.1% less variability and only 2–3 W
PV Systems with Machine Learning: standard deviation at the GMPP.
A Neuro-Fuzzy MPPT Control for
PSCs with Uncertainties. Electronics Keywords: maximum power point tracking; machine learning; partial shading; terminal sliding
2023, 12, 1720. https://doi.org/ mode control
10.3390/electronics12071720

Academic Editor: Enrique

Romero-Cadaval
1. Introduction
Received: 1 February 2023 The continuous increase in global warming and the decrease in fossil energy sources
Revised: 10 March 2023
has led to a sharp inclination towards renewable sources as a substitute source of energy.
Accepted: 28 March 2023
Among these renewable sources, the solar system has been extensively used for power
Published: 4 April 2023
generation in a variety of applications due to its multiple benefits, such as uninterrupted
power, no noise, no pollution, and easy maintenance. This increasing demand for power
generation using photovoltaic (PV) systems for both residential and industrial areas re-
Copyright: © 2023 by the authors.
quires an appropriate and efficient optimization of energy production systems. To draw
Licensee MDPI, Basel, Switzerland. maximum PV energy, many methods have been developed in the past, such as incremental
This article is an open access article conductance, perturb and observe, constant voltage, parasitic capacitance, and constant
distributed under the terms and current. This allows the controller to harvest the maximum available power from the PV
conditions of the Creative Commons system under varying solar irradiance and temperature scenarios; these controllers are
Attribution (CC BY) license (https:// usually known as maximum power point tracking (MPPT) controllers. However, most
creativecommons.org/licenses/by/ MPPT methods suffer from a lack of strict convergence analysis and are not capable of
4.0/). handling partial shading conditions (PSCs) [1,2].

Electronics 2023, 12, 1720. https://doi.org/10.3390/electronics12071720 https://www.mdpi.com/journal/electronics

Electronics 2023, 12, 1720 2 of 22

1.1. Literature Review

This research aims to find a more effective solution for PV systems that are partially
shaded by analyzing and modeling their performance. By simulating the relationship
between environmental conditions and the PV array’s output characteristics, we can gain a
deeper understanding of the special effects of PS. This knowledge is crucial for developing
an MPPT algorithm. Power generation in solar systems is influenced by the intermittence
of both temperature and irradiance. The output power can also fluctuate because of the
non-linear relationship between current and voltage in conventional PV cells. Therefore,
incorporating an MPPT algorithm can help to maximize power output under varying
meteorological conditions. However, when PV systems are partially shaded, they exhibit
distinct characteristics, and multiple peaks may appear on the P-V characteristic curve.
The impact of shading depends on the pattern and placement of the PV arrays, which can
decrease the efficiency of the tracking algorithm as the PV array tends to operate at a local
MPP [3–5].
MPPT calculations are an essential part of the operation of PV systems. These calcula-
tions are designed to optimize the output power of PV panels by determining the point
on the P-V curve where maximum power can be obtained. Under ideal conditions where
all PV cells receive the same amount of sunlight, the MPPT algorithm can quickly and
easily locate a single peak on the P-V curve and adjust the system accordingly. However,
there are scenarios where the P-V curve of the panel has multiple local peaks. One of the
prevalent reasons is PS, which happens when the cells in a panel are exposed to varying
levels of sunlight. This can occur when the panel is partially blocked by an object such as a
tree or building, or when the panel is installed in a location with uneven sunlight. In such
scenarios, the standard MPPT calculations may only identify the local maximum power
point (LMPP) in the previous working point region, as the global maximum power point
(GMPP) could be located much farther away on the P-V curve [6]. Therefore, it is crucial to
utilize sophisticated MPPT methods and techniques to trace the global peak power point
during PS scenarios and augment the general effectiveness and operation of the PV system.
This will immensely reduce the competitiveness of the PV panel, in particular during PS.
To solve PS, numerous approaches have been proposed to alleviate the effect of PSCs in
such PV panels and a lot of research has been carried out, focusing on finding the GMPP
by reducing the search area.
Hu et.al presented an idea that states the current of faulted cells or modules increases
under the PSC, causing the temperature rise of a few faulted modules or cells [7]. Therefore,
there are a lot of LMPPs. The usage of a thermal camera to identify PSCs is also proposed.
The proposed approach can detect the cell or module faults and can make use of the thermo-
graphical data gathered from panels to split the PV array into healthy and unhealthy
segments and also efficiently determine PS. However, this method consumes a lot of
computational time for global maximum power point tracking (GMPPT) [8].
Tamir Shaqarin suggested an approach that works well for tracking precision and
steady-state error to track the GMPP under any climate condition. The proposed approach is
using “particle swarm optimization (PSO) through targeted position-mutated elitism” (PSO-
TPME) with a reinitialization mechanism on a PV system under partial shading conditions.
The fast-converging and global exploration capabilities of PSO-TPME make it appealing
for online optimization. But a significant implementation complexity is associated with
PSO-TPME based MPPT [9].
A MPPT approach with minimal complexity is suggested by A. Safari which is founded
on an adjustable step size incremental conductance method and a straightforward linear
equation [10,11]. The aim of this approach is to relocate the operating point near the
GMPP. This technique is based on a variable step size incremental conductance, which
automatically adjusts the step size to track the GMPP and minimize energy loss. However,
this method has a significant drawback, as the linear function may not be effective when
the PV array has multiple LMPPs. Furthermore, this approach may struggle to adapt to
system parameters that change over time.
Electronics 2023, 12, 1720 3 of 22

Different “perturb & observe” (P&O) and incremental conductance (InC) MPP tech-
niques have been used in the PV system by the researcher for detecting the GM using P-V
& I-V curves. The techniques used in [12] are evaluated based on speed, accuracy, and
complexity. Fractional short circuit current (FSCC), P&O, fractional open circuit voltage
(FOCV), and InC techniques are conventional and the most commonly employed tech-
niques in PV energy systems. However, the major drawback of these techniques is lack of
robustness and frequent oscillations around local optima [13].
An adaptive neuro-fuzzy (NF) interference system strategy is proposed in [14,15]
which is applied to extract MMPT for the PV system under intermittent environmental
circumstances; however, this technique based on machine learning requires a huge database.
Additionally, this technique suffers from the chattering effect, steady-state error, and
oscillations in the desired output.
Other researchers have also employed population-based global optimization methods
which have been combined with deep neural networks to enhance their global exploration
capabilities and reduce their computational complexity [16–18] such as the firefly algorithm
(FA) [19], the artificial bee colony algorithm (ABC) [20], and the genetic algorithm (GA) [21]
to achieve maximum power, but these algorithms suffer from convergence speed, require
lot of tuning parameters (population size, crossover probability, and mutation rate), and are
sentinel to noise. Moreover, all these PSO, GA, FA, and ABC population-based algorithms
are not efficient when it comes to the control problem, because of their inability to handle
uncertainty and nonlinear systems efficiently and are less flexible to changing environment
conditions.
In conclusion, it is obvious that there are multiple MPPT techniques available with
distinct features, including the type of sensing material, rate of convergence, level of
intricacy, efficiency, expense, and suitability, all targeting to locate the GMPP under PSCs.
Each method is associated with its advantages and disadvantages but a general drawback
that is being observed in all traditional techniques is the challenge of chattering and slow
convergence. Focusing on this research problem, a hybrid MPPT technique is proposed that
can perform under varying climate conditions and shading patterns with slight chattering,
and can assure fast and finite convergence to GMMP.

1.2. Original Contribution

The following are in-depth descriptions of the key innovations and significant contri-
butions made through this research.
1. This research presents a strong and sophisticated controller known as the nonlinear
terminal sliding mode controller (TSMC) that is specifically designed to track the
MPPT of PV arrays’ PSCs by utilizing a non-inverting buck-boost converter.
2. To achieve this, the proposed controller utilizes a neuro-fuzzy network (NFN) for
reference generation, which is trained using over 22,000 distinct PS scenarios.
3. The proposed controller is designed to ensure fast and finite-time convergence, pro-
viding a reliable and efficient solution for MPPT under PSCs.
4. The robustness and chattering minimization around the GMPP are tested by introduc-
ing uncertainties in the system, demonstrating the success of the proposed controller
in challenging operating conditions.
5. To demonstrate the efficacy of the proposed controller, an experimental setup is estab-
lished, which allows for a comprehensive evaluation of the controller’s performance.
6. To evaluate its performance, a comparison of the proposed controller with other
algorithms is already available. The results are presented in Table 1, which illustrates
the superiority of the proposed controller in terms of its performance characteristics.
The comparison highlights the effectiveness of the proposed controller and its ability
to perform better than other existing algorithms.
 Electronics 2023, 12, x FOR PEER REVIEW 4 of 22

6. To evaluate its performance, a comparison of the proposed controller with other al‐
Electronics 2023, 12, 1720
gorithms is already available. The results are presented in Table 1, which illustrates4 of 22
the superiority of the proposed controller in terms of its performance characteristics.
The comparison highlights the effectiveness of the proposed controller and its ability
to perform better than other existing algorithms.
Table 1. Comparison of proposed work with existing techniques.
Table 1. Comparison of proposed work with existing techniques.
References Methodology Merits Demerits
References Methodology
Incremental Merits
Fast response and Demerits
Offline and large
Punitha et. al. [22] Incremental Fast response and Offline and large time
Punitha et. al. [22] conductance good tracking computational
conductance good tracking computational time
Patel et. al. [23] Neural networks High tracking speed Oscillations at MPPT
Patel et. al. [23] Neural networks High tracking speed Oscillations at MPPT
Particle swarm Reduce steady Large
Koad et. al. [24] Particle swarm Reduce steady state Large computational
Koad et. al. [24] optimization state error computational time
optimization error time
Slow Slow tracking and
tracking and
El-Helw et. al. [25] Perturb and observe Easy implementation oscillation
El‐Helw et. al. [25] Perturb and observe Easy implementation oscillation around
around MPTT
MPTT
Robust, fast
Neuro-fuzzy Robust, fast Offline technique,
Proposed work Neuro‐fuzzy and convergence and Offline technique,
Proposed work and TSMC convergence and requires large dataset
TSMC minimum chattering requires large dataset
minimum chattering

The presentstudy
The present studyisisorganized
organizedinin a systematic
a systematic manner
manner to comprehensively
to comprehensively address
address
the
the proposed PV system with a MPPT controller. The following sections outline the meth‐ the
proposed PV system with a MPPT controller. The following sections outline
methodology and results
odology and results of the of the study.
study.
Section
Section 22 delves
delvesinto intothe
thestatistical
statistical
and and mathematical
mathematical modeling
modeling ofgiven
of the the given PV sys-
PV sys‐
tem,
tem, providing
providing aathorough
thoroughunderstanding
understanding of of
thethe system’s
system’s behavior.
behavior. The The shading
shading effect,
effect, a a
crucial aspect of
crucial aspect ofPVPVsystems,
systems,isisdescribed
described in in Section
Section 3. Section
3. In In Section 4, use
4, the the of
usea of a machine
machine
learning-based neuro-fuzzynetwork
learning‐based neuro‐fuzzy network(MLNFN)
(MLNFN)isispresented
presentedfor forthe
the generation
generation andand training
train‐
of
ingreference
of reference voltage.
voltage.ThisThismethodology
methodology is is applied
appliedtotoimprove
improve thethe performance
performance of the of the
MPPT
MPPT controller.
controller. The Theaverage
averagestate‐space
state-space model
model of of
thethe DC—DC
DC—DC buck-boost
buck‐boost convertor
convertor is is
explained in Section 5, while
explained while Section
Section 66 presents
presentsthe therobust
robustnonlinear
nonlinearTSMC.TSMC.To Toevaluate
evaluate the
the proposed controller’s performance under varying environmental
proposed controller’s performance under varying environmental conditions, simulation conditions, simula‐
tion results
results are analyzed
are analyzed using
using MATLAB/Simulinkin
MATLAB/Simulink inSections
Sections 7 andand 8. 8.The
Thehardware
hardware val‐
valida-
idation of the proposed system and a comprehensive performance
tion of the proposed system and a comprehensive performance analysis are presented in analysis are presented
in Sections
Sections 9 and
9 and 10,10,respectively.
respectively.Finally,
Finally,the
thestudy
studyconcludes
concludes in in Section
Section10,10,summarizing
summarizing the
key findings and highlighting the contributions made to the field of PVPV
the key findings and highlighting the contributions made to the field of systems
systems and
and MPPT
MPPT controllers.
controllers.

2. PV
2. PV System
SystemMathematical
MathematicalModeling
Modeling
The PV
The PV cell
cell has
hasaap‐n
p-njunction
junctionandandproduces
produceselectric power
electric duedue
power to the photons.
to the It
photons. It
consists of a current source I ph, a series resistance Rs, a shunt resistance Rp, and a diode, as
consists of a current source Iph , a series resistance Rs , a shunt resistance Rp , and a diode, as
shown in
shown in Figure
Figure 1.1.

Figure 1. PV
Figure PV array
arrayequivalent
equivalentmodel.
model.

To evaluate the parameters of the photovoltaic (PV) system it is necessary to have

knowledge of the PV power-voltage or current-voltage curve under standard conditions
of measurement (SCM). This information can be obtained through the testing and char-
acterization of the PV module under SCM, which typically includes a specific irradiance
level and cell temperature. This information can then be used to model the PV system
and determine the performance and efficiency of the system under different operating
conditions.
Electronics 2023, 12, 1720 5 of 22

As Rp and Rs have low values, they can be neglected in some cases. This means that
the total current is mainly determined by the photocurrent and the resistance losses are
considered to be negligible. However, this will depend on the specific PV system and the
operating conditions, and it is always important to consider the actual values of Rp and Rs
to make sure they can be neglected. The current of a PV array arranged in a combination of
series and/or parallel can be described by [26,27].
    
NS NS
  
V + RS N P
I V + R S N I
I = IPV NP − IO NP exp  − 1 −  P , (1)
VT aNs RP N NS
P

where
IPV : PV array current
IO : Diode saturation current
a: Ideal factor
RS : Resistance in series
R P : Resistance in parallel
NS : Number of series cell
NP : Number of parallel cell
VT : Thermal voltage
PV array thermal voltage is given by,

NS KT
VT = (2)
q

where
q: Electron charge
K: Boltzmann’s constant
T: Temperature of p-n junction
The PV current IPV is given by,

G
IPV = ( IPV N + Ki ∆T ) (3)
GN

where
IPV N : Nominal condition PV current
G: Irradiance at the surface of panel
GN : Nominal condition irradiance
Ki : Temperature coefficient of short circuit current
∆T: Difference of nominal and actual temperature
Saturation current of the diode is represented by,

ISCN + Ki ∆T
IO = V  (4)
ocn+Ki ∆T
exp aVt −1

where
ISCN : Nominal condition short circuit current
VOCN : Nominal condition open circuit voltage
MATLAB/Simulink is used to model and simulate the above-mentioned equations
and corresponding results are presented in Figures 2, 3, 9 and 10.
 Electronics 2023, 12, x FOR PEER REVIEW 6 of 22
Electronics 2023, 12, x FOR PEER REVIEW 6 of 22

Electronics 2023, 12, 1720 6 of 22

MATLAB/Simulink
MATLAB/Simulink is is used
used to
to model
model and
and simulate
simulate the
the above‐mentioned
above‐mentioned equations
equations
and
and corresponding results are presented in Figures 2, 3, 9 and 10.
corresponding results are presented in Figures 2, 3, 9 and 10.

Figure 2.
Figure 2. PV
2. PVcurve
PV curveat
curve atatdifferent
differentirradiances.
different irradiances.
irradiances.

Figure 3. PV curve at different temperatures.

Figure 3.
Figure 3. PV
PVcurve
curveatatdifferent
differenttemperatures.
temperatures.
These
These figures
figures illustrate the I‐V and P‐V attributes of the PV system, which are used
These figures illustrate
illustratethe theI‐V
I-VandandP‐VP-Vattributes
attributesofofthethePV PVsystem,
system,which
whichareare
used
used in
in the current work and subject to different levels of irradiances and temperatures. Four
the current work and subject to different levels of irradiances and temperatures.Four
in the current work and subject to different levels of irradiances and temperatures. Four PV
PV
PV arrays
arrays are linked in aa series such that each array comprises of fifteen series modules
arrays are are linked
linked in ainseriesseries such
such thatthat
eacheach array
array comprises
comprises ofof fifteen
fifteen seriesmodules
series modulesand a
and
and a
a total
total output
output of
of 24
24 KW
KW power
power of
of the
the other
other two
two parallel
parallel strings.
strings.
total The
output of 24 KW power ofisthe other two parallel strings.
The P‐V
The P‐V
P-V
curve
curve
curve
of
ofofthe
the
the
system
system
system
presented
is is
presented
presented
in
in Figure
Figure
in
22 with
Figure with aa constant
2 with constant
acurve
temperature
temperature
constant
that
that that
temperature
is
is directly proportional to power. While Figure 3 shows the P‐V curve of the system at
directly proportional to power. While Figure 3 shows the P‐V of the system at
is directly proportional
different to power.irradiance
While Figure 3 shows theinversely
P-V curve of the system at
different temperatures
temperatures by by keeping
keeping irradiance constant,
constant, which
which is is inversely proportional
proportional to
to
different
power. temperatures by keeping irradiance constant, which is inversely proportional to
power.
power.
3. Influence of Shading Effect on PV Array
3. Influence of Shading Effect on PV Array
3. Influence of Shading Effect on PV Array
It
It is
is aa common
common practice
practice to
to connect
connect multiple
multiple solar
solar panels
panels in
in aa series
series or
or parallel
parallel config‐
config‐
It is
uration ameet
common practice to connect multiple solar panels in a blocked
series or parallel con-
uration to meet power requirements. However, when certain panels are blocked from
to power requirements. However, when certain panels are from sun‐
sun‐
figuration
light to meet power requirements. However, when certain panels are blocked from
light by
by passing
passing clouds,
clouds, nearby
nearby trees
trees oror buildings,
buildings, the
the shaded
shaded cells
cells absorb
absorb some
some ofof the
the
sunlight
power by passing clouds, nearby trees or buildings, the shaded cells absorb some of the
power generated
generated by by unshaded
unshaded cells,
cells, resulting
resulting in
in the
the generation
generation ofof heat
heat which
which can
can cause
cause
power
damage generated by unshaded cells, resulting in the generation of heat which can cause
damage to to the
the shaded
shaded cells
cells [28].
[28].
damage to the shaded cells [28].
To prevent this, bypass diodes can be employed as depicted in Figure 4, which help to
prevent a negative voltage across the shaded cells. From Figure 4, the bypass diode begins
to conduct when the condition Vi − ∑in=1 Vi ≥ VD is satisfied where i 6= 2 and VD is the
forward voltage drop of the diode [29,30].
 to prevent a negative voltage across the shaded cells. From Figure 4, the bypass diode
Electronics 2023, 12, x FOR PEER REVIEW 7 of 22
begins to conduct when the condition 𝑉 ∑ 𝑉 𝑉 is satisfied where i ≠ 2 and 𝑉 is
the forward voltage drop of the diode [29,30].
However, activating diodes at different voltage levels alters the characteristics of the
To prevent
PV system this, bypass
and results diodespeaks,
in multiple can beas employed
shown inasFigure
depicted in Figure
5. This 4, which
trait can causehelp
the7 of 22
Electronics 2023, 12, 1720
to prevent
system a negative
to operate voltage
at a local peakacross theofshaded
instead cells.global
the optimal Frompeak,
Figure 4, the bypass
resulting diode
in a decrease
begins
in to conduct when the condition 𝑉 ∑
PV efficiency. 𝑉 𝑉 is satisfied where i ≠ 2 and 𝑉 is
the forward voltage drop of the diode [29,30].
However, activating diodes at different voltage levels alters the characteristics of the
PV system and results in multiple peaks, as shown in Figure 5. This trait can cause the
system to operate at a local peak instead of the optimal global peak, resulting in a decrease
in PV efficiency.

Figure 4. Bypass
Figure Bypassdiode
diodeparallel
parallelwith
withPV
PVcells.
cells.

However, activating diodes at different voltage levels alters the characteristics of the
PV system and results in multiple peaks, as shown in Figure 5. This trait can cause the
system to operate at a local peak instead of the optimal global peak, resulting in a decrease
Figure
in 4. Bypass diode parallel with PV cells.
PV efficiency.

Figure 5. Global and local peaks due to the shading effect.

4. Machine Learning Based Proposed System Framework

Figure 6 illustrates the MLNF technique used in this research, which is based on the
Takagi‐Sugeno (TS) fuzzy inference system. TS and Mamdani are two popular types of
Figure inference
Figure
fuzzy 5. Global
Globalandand local
localpeaks
systems peaksdue
used induetoto
the thethe
fieldshading
shading
of fuzzyeffect.
effect.
logic. The main difference between the
two is the way they handle fuzzy rules and fuzzy outputs. In Mamdani‐type systems, the
4.
4. Machine LearningBased
Machine Learning BasedProposed
Proposed System
System Framework
Framework
output of each rule is a fuzzy set, which is then combined to form the final output using
Figure 6 illustrates
6 illustratessuch
fuzzy logic operations the MLNF
the MLNF
as union technique
technique
or centroid used
used in this
in this research,
research,
defuzzification. In whichwhich is based
is based
contrast, TS‐typeonsys‐ on the
the
Takagi-Sugeno
Takagi‐Sugeno
tems use a linear (TS)
(TS) fuzzyinference
fuzzy
combination inference system.
of the system.
inputs TS and
toTSgenerate
and Mamdani
Mamdani are are
twotwo
a crisp output. popular
popular
There aretypes types
severalof of
fuzzy inference
fuzzy inference
advantages systems
systems
of using used
used
TS‐type ininthe
systemsthefield ofof
field
as theyfuzzy
fuzzy
are logic. TheThe
logic. main
computationally difference
main between
difference
efficient, between
accurate, the the
pro‐
two is
two
vide the
theway
better way they
theyhandle
interpolation, handle
andfuzzy
are rules
fuzzy
easy and
rules
to fuzzy
and
tune outputs.
fuzzy
compared toIn Mamdani‐type
outputs. systems,
In Mamdani-type
Mamdani‐type systems the
systems,
[31].
output
the of each
output rule is
of each a fuzzy
rule set, which
is a fuzzy set, is then combined
which to form the
is then combined to final
formoutput using
the final output
fuzzy fuzzy
using logic operations such as union
logic operations such as or centroid
union ordefuzzification. In contrast, TS‐type
centroid defuzzification. sys‐ TS-
In contrast,
temssystems
type use a linear
use combination of the inputs
a linear combination to generate
of the inputs toa generate
crisp output. There
a crisp are several
output. There are
advantages
several of usingof
advantages TS‐type systems systems
using TS-type as they are as computationally efficient, accurate,
they are computationally efficient, pro‐
accurate,
vide better
provide interpolation,
better interpolation,and and
are easy to tune
are easy compared
to tune comparedto Mamdani‐type
to Mamdani-type systems [31]. [31].
systems
The proposed framework has five inputs, including four different irradiance values
(G1, G2, G3, and G4) and one temperature (T) parameter. The fuzzification layer, which is
the input layer, consists of three triangular membership functions for each parameter. The
output layer comprises a linear equation for each rule. The MLNFN generates a reference
voltage for the peak power from the PV arrays under PS and varying environmental
conditions.
Electronics 2023,
Electronics 12,12,
2023, 1720
x FOR PEER REVIEW 8 of 228 of 22

Figure 6.
Figure 6. Proposed
ProposedMLNF
MLNFcontrol
controlscheme.
scheme.

The proposed
Large datasetsframework
are essentialhas for
fivetraining
inputs, including
the MLNFN.four different irradiance
These datasets arevalues
generated
(G1, G2, G3, and G4) and one temperature (T) parameter. The fuzzification
by modeling the PV array system in MATLAB/Simulink under different (PSCs). layer, which
To train
is the
the input
NFN, layer, 22,000
around consistsscenarios
of three triangular
of PSCs aremembership functions for
randomly generated foreach
MPPparameter.
voltages to be
The output
used layer comprises
as reference voltages fora linear equation for each rule. The MLNFN generates a ref‐
the controller.
erence voltage for the peak power from the PV arrays under PS and varying environmen‐
Figure 7 illustrates the reference voltages generated using the MLNFN against three
tal conditions.
different membership functions at shading pattern SP1, which lasts from 0 to 0.5 s, and
Large datasets are essential for training the MLNFN. These datasets are generated by
SP2, which lasts from 0.5 to 1 s. The voltage changes abruptly from 1645 V to 1265 V
modeling the PV array system in MATLAB/Simulink under different (PSCs). To train the
when irradiances change from SP1 to SP2 in the case of triangular membership functions.
NFN, around 22,000 scenarios of PSCs are randomly generated for MPP voltages to be
Table 2 provides a comparative analysis of reference voltages generated using different NF
used as reference voltages for the controller.
Electronics 2023, 12, x FOR PEER REVIEW 9 of 22
techniques.
Figure 7The table helps
illustrates in understanding
the reference the effectiveness
voltages generated using theof the proposed
MLNFN against technique
three
over the existing
different membershipones.functions at shading pattern SP1, which lasts from 0 to 0.5 s, and
SP2, which lasts from 0.5 to 1 s. The voltage changes abruptly from 1645 V to 1265 V when
irradiances change from SP1 to SP2 in the case of triangular membership functions. Table
2 provides a comparative analysis of reference voltages generated using different NF tech‐
niques. The table helps in understanding the effectiveness of the proposed technique over
the existing ones.

Table 2. Comparison of reference voltage prediction for various NF techniques.

FIS Membership Shading Pattern Reference Voltage Actual Voltage

Error (V)
Functions (SP) (V) Generated (V)
TRIMF SP1 1262 1262.1 0.1
TRAPMF SP1 1262 1235 27
GUASSMF SP1 1262 1258 4

Figure 7.
Figure 7. Reference
Referencevoltage
voltagegeneration
generationusing various
using NFNF
various techniques.
techniques.

5. Non‐Inverting
Table Average
2. Comparison State‐Space
of reference voltage Model of Buck‐Boost
prediction Converter
for various NF techniques.
The DC—DC buck‐boost converter in non‐inverting configuration moves up or
FIS the
down Membership Shading
voltages to force Pattern
the PV Reference
array to operate Actual
at the MPP, fromVoltage
input (PV array)
Errorto
(V)
Functions (SP) Voltage (V) Generated (V)
output (load). With the help of the switching period T, the converter is periodically con‐
trolled where:
TRIMF T = Ton + Toff, Ton SP1
is the ON time and1262
Toff is the OFF time. The converter’s duty
1262.1 0.1
ratio isTRAPMF
defined by u = Ton/T. SP1 1262 1235 27
To reduce the waves in the converter, input voltage capacitor C1 is used; while 4for
GUASSMF SP1 1262 1258
limiting the output voltage capacitor, C2 is used. In this work, it is assumed that the con‐
verter is operating in continuous conduction mode (CCM).
Figure 8 shows the approximate circuit of the non‐inverting buck‐boost converter
with two switching intervals. The first switching interval has both switches, 𝑆 and 𝑆
active while the diodes 𝐷 and 𝐷 are inactive. In the second switching interval, both the
diodes 𝐷 and 𝐷 are active while the switches; 𝑆 and 𝑆 are inactive [32,33]. In the
first interval, according to Kirchhoff’s current and voltage law, we have:
Electronics 2023, 12, 1720 9 of 22

5. Non-Inverting
Figure Average
7. Reference voltage State-Space
generation ModelNF
using various oftechniques.
Buck-Boost Converter
The DC—DC buck-boost converter in non-inverting configuration moves up or down
5. Non‐Inverting
the voltages to force Averagethe PV State‐Space Model of
array to operate at Buck‐Boost
the MPP, from Converter
input (PV array) to output
(load).TheWithDC—DC
the help buck‐boost converterperiod
of the switching in non‐inverting configuration
T, the converter moves up
is periodically or
controlled
down the
where: T =voltages
Ton + Toffto, T
force
on is the
the PVONarray
time to operate
and Toff is at
thethe MPP,
OFF from
time. Theinput (PV array)
converter’s to ratio
duty
output
is defined(load).
by uWith
= Tonthe /T.help of the switching period T, the converter is periodically con‐
trolled
Towhere:
reduceT the
= Tonwaves
+ Toff, Tin
on is theconverter,
the ON time and Toff is
input the OFFcapacitor
voltage time. TheCconverter’s duty
1 is used; while for
ratio is defined
limiting by u =voltage
the output Ton/T. capacitor, C2 is used. In this work, it is assumed that the
To reduce
converter the waves
is operating in the converter,
in continuous input voltage
conduction capacitor C1 is used; while for
mode (CCM).
limiting the output voltage capacitor, C2 is used. In this work, it is assumed that the con‐
Figure 8 shows the approximate circuit of the non-inverting buck-boost converter with
verter
two is operating
switching in continuous
intervals. The firstconduction
switching mode (CCM).
interval has both switches, S1 and S2 active
Figure 8 shows the approximate circuit of the non‐inverting buck‐boost converter
while the diodes D1 and D2 are inactive. In the second switching interval, both the diodes
with two switching intervals. The first switching interval has both switches, 𝑆 and 𝑆
D1 and D2 are active while the switches; S1 and S2 are inactive [32,33]. In the first interval,
active while the diodes 𝐷 and 𝐷 are inactive. In the second switching interval, both the
according to Kirchhoff’s current and voltage law, we have:
diodes 𝐷 and 𝐷 are active while the switches; 𝑆 and 𝑆 are inactive [32,33]. In the
first interval, according to Kirchhoff’s current
I = and voltage
I − I law, we have: (5)
PV PV L
𝐼 𝐼 𝐼 (5)
VL = VC1 (6)
𝑉 𝑉 (6)

𝑉VC2
𝐼 IC2 = − (7) (7)
𝑅R

Figure 8.
Figure 8. Non−inverting
Non−invertingconfiguration
configurationof of
DC‐DC buck‐boost
DC-DC converter.
buck-boost converter.

Whereas in
Whereas inthe
thesecond
second interval 𝑆 Sand
interval 𝑆 are inactive and 𝐷 and 𝐷 are forward‐
1 and S2 are inactive and D1 and D2 are forward-
biased. Using Kirchhoff’s current and voltage
biased. Using Kirchhoff’s current and voltage law, wewe
law, have:
have:

IC1 = IPV (8)

VL = −VC2 (9)

VC2
IC2 = IL − (10)
R
In light of the volt second balance of the inductor and charge balance of the capacitor,
we can express:
dvc1 I I
= PV − L u (11)
dt C1 C1

di L V V
= C1 u − C2 (1 − u) (12)
dt L L

dvc2 I V
= L (1 − u) − C2 (13)
dt C2 RC2
Electronics 2023, 12, 1720 10 of 22

Averaging the model for one switching duration and considering x1 , x2 and x3 to be
the average value of VC1 , IL and VC2 , we can write, x1 =< vc1 >, x2 =< i L >, x3 =<
vC2 > and u =< u >.
Hence Equations (11)–(13) can be rearranged as,

. i pv x
x1 = − 2u (14)
c1 c1

. x1 x3
x2 = u− (1 − u ) (15)
L L1
. x2 x
x3 = (1 − u ) − 3 (16)
c2 RC2
The final Equations (14)–(16) are utilized in the designing of PV system control law.

6. Design of Terminal Sliding Model Control

A controller design based on the nonlinear robust terminal sliding mode (TSM) is
proposed for tracking the MPPT of PV arrays under PSCs using a non-inverting buck-boost
converter. In this controller, the error, e1, is defined as the discrepancy between the desired
output voltage of the PV array and the actual one and is given in Equation (17). The
controller uses this error to adjust the reference voltage generated by the MLNFN algorithm
and to extract the maximum power from the PV array,

e1 = x1 − x1d (17)

where x1 is Vpv and x1d refers to Vre f . The derivative of Equation (17) with its dynamics
reported in (15) becomes,
. i pv x
e1 = − 2 u − x1d (18)
c1 c1
The first stage is to design a sliding surface and the next stage is the selection of a
control law for holding the system trajectory on the sliding surface making the tracking
error zero. The equation of TSMC is given by,
  n −1
d
s= +∝ e1 (19)
dt

s = e2r + ∝ e1 (20)
.
where e2 = n represents the relative system degree, e1 is the error among the ∝, desired
e1 ,
reference voltage, a positive parameter chosen randomly or by some approach whichever
is more suitable choice and the output voltage, r is rational power equal to p/q.
By taking the derivative of sliding surface we have,
. . .
s = re2r−1 e2 + ∝ e1 (21)

Control law is given by Equation (22),

u(t) = ueq + udis (22)

udis = −k1 (s) − k2 sign(s) (23)

where ueq is the equivalent control vector while udis is the discontinuous control (the
correction factor) vector which is given by udis = ksign(s), where k is a controlled gain.
To obtain ueq , Equation (22) will be simplified, the parameters used in the equation are
given as,
e1 = x1 − x1d (24)
Electronics 2023, 12, 1720 11 of 22

. i pv x
e1 = − 2 u − x1d (25)
c1 c1
1 1
e2 = ∝ r e1 r (26)
Taking derivative of e2 yields,

. 1 1 1 −1 .
e2 = ∝ r e1 r e1 (27)
r
By substituting the parameters in Equation (22) we obtain,

.
h 1 1 i r −1 h 1 1  .
i
i ux2
s = r ∝ r e1 r ∝ r e1 r cpv − c1 − x1d +

i .
 1 (28)
∝ cpv − ux c
2
− x 1d
1 1

.
By taking s = 0 we obtain, ueq as,
.
i pv (r − 1) + x1d c1 (1 − r )
ueq = (29)
x2 (r − 1)

Finally, by adding Equations (23) and (29) we obtain the control law u(t),
.
i pv (r − 1) + x1d c1 (1 − r )
u(t) = − k1 k2 sign(s) (30)
x2 (r − 1)

7. Proposed Control System Performance Validation

This section explains the validation of the proposed system for MPPT under various
PSCs using MATLAB/Simulink.

7.1. Simulation Setttings

MATLAB 2017Ra is used to perform the simulations where powergui and a constant
time solvent are opted for to run the simulation. The information about the PV array being
used in this research is given in Table 3. The parameter values used in converter and
controller designing are given in Table 4. Three distinct shading patterns (SPs) namely SP1,
SP2, and SP3 are subjected to varying irradiance, while SP4 exhibits uniform irradiance, as
mentioned in Table 5. For the SP4 case with regular irradiance, only a single PV curve is
produced, which in turn keeps a simpler detection for the MPP through any traditional
methods. However, in the remaining three cases of PS, it is difficult to identify the global
peak, as the PV characteristic curve now exhibits multiple local maxima points instead of
just one GMPP. These local maxima points appear due to the shading of certain portions of
the PV array, which can cause the power output of the shaded portions to decrease, creating
multiple peaks in the PV curve, making it challenging to pinpoint the GMPP accurately.

Table 3. Parameters of PV system.

Parameters Symbols Values

Maximum power PMAX 200 W
Open circuit voltage VOC 32.90 V
Optimum voltage VMP 26.30 V
Short circuit current ISC 8.210 A
Optimum current IMP 7.610 A
Temperature coefficient of ISC TSC 0.00318 A/◦ C
Temperature coefficient of VOC TOC −0.123 V/◦ C
Parallel resistance RP 601.33 Ohms
Series resistance RS 0.23 Ohms
Electronics 2023, 12, 1720 12 of 22

Table 4. Parameters of converter and controller.

Parameters Symbols Values

Capacitor at input C1 1 mF
Capacitor at output C2 48 µF
Inductor L 20 mH
Switching frequency FS 5000 Hz
Constant K1 10
Constant K2 100
Rational power R 0.501

Table 5. SP for P-V and I-V curves.

SP G1 (w/m2 ) G2 (w/m2 ) G3 (w/m2 ) G4 (w/m2 )

SP1 200 400 600 1000
SP2 400 500 800 800
SP3 600 600 1000 1000
SP4 1000 1000 1000 1000

The proposed MLNFN TSMC strategy is evaluated from three distinct characteristics
i.e., (1) robustness to environmental changes (2) PSCs, and (3) controller uncertainties.
Initially, the simulations were carried out to achieve the P-V and I-V attribute curves
Electronics 2023, 12, x FOR PEER REVIEW
of the four SPs mentioned above in Table 5. Figure 9 shows the P-V curve 13 ofattributes
22 under
Electronics 2023, 12, x FOR PEER REVIEW 13 of 22
SP1 to SP4 while Figure 10 shows I-V curve under SP1 to SP4.

Figure 9. P‐V curves at different SPs.

Figure
Figure 9.9. P-V
P‐V curves
curves at different
at different SPs. SPs.

Figure 10.
10. I‐V
Figure
Figure 10.I‐VI-V
curves at
curves atdifferent
curves differentSPs.
SPs.
at different SPs.
7.2. Test
7.2. Test Case
Case Senarios
Senarios
Two different
Two different sets
sets of
ofSPs
SPswith
withdifferent
differenttemperatures
temperaturestaken
takenfor
forthe
thecomparative
comparative per‐
per‐
formance analysis
analysis of
of the
theproposed
proposedtechnique
techniqueare
areshown
shownininTable
Table6.6.

Table 6. SP
SP for
for test
test case
casescenarios.
scenarios.
Electronics 2023, 12, 1720 13 of 22

It can be noted from the results that irradiances are directly proportional to the cur-
rent/voltage; however, when SPs are different then various local peaks have happened.

7.2. Test Case Senarios

Two different sets of SPs with different temperatures taken for the comparative perfor-
mance analysis of the proposed technique are shown in Table 6.

Table 6. SP for test case scenarios.

SP G1 (w/m2 ) G2 (w/m2 ) G3 (w/m2 ) G4 (w/m2 ) T (◦ C)

SP1 200 400 600 1000 30
SP2 400 500 800 800 25

It was observed in the case of the SP1 scenario that the GMPP was situated at 1645 V
and the associated output power was 15,000 W, as shown in Figure 11. While for the case of
SP2, Figure 12 shows that MPP was at 1250 V and the associated output power was 8000 W.
Electronics
Electronics 2023,
2023, 12,
12, xx FOR
FOR PEER
PEER REVIEW
REVIEW 14
14 of
These results clearly show that the proposed MLNFN TSMC efficiently tracks the of 22
GMPP
22 in
the existence of local peaks.

Figure
Figure 11.
Figure 11. Voltage
Voltageand
Voltage andpower
and powerat
power atatSP1.
SP1.
SP1.

Figure
Figure 12.
Figure 12. Voltage
Voltageand
Voltage andpower
and powerat
power atatSP2.
SP2.
SP2.

8.
8. Comparative
Comparative Analysis
Analysis of
of Proposed
Proposed Technique
Technique
For
For detailed analysis regarding the performance
detailed analysis regarding the performance improvement
improvement achieved
achieved with
with the
the
proposed
proposed strategy, a comparative study with existing conventional techniques was
strategy, a comparative study with existing conventional techniques was car‐
car‐
ried
ried out
out and
and can
can be
be found
found in
in the
the subsequent
subsequent sections.
sections.

8.1.
8.1. Performance
Performance Analysis
Analysis without
without Uncertainities
Uncertainities
Electronics 2023, 12, 1720 14 of 22

8. Comparative Analysis of Proposed Technique

For detailed analysis regarding the performance improvement achieved with the
proposed strategy, a comparative study with existing conventional techniques was carried
out and can be found in the subsequent sections.

8.1. Performance Analysis without Uncertainities

Initially, the proposed MPPT technique’s efficiency was evaluated under both steady-
state and dynamic conditions. The dynamic response of the technique is depicted in
Figure 13, which shows the tracking of voltage when there was a sudden change in the set
point from SP1 to SP2. The PV terminal voltage at 1645 V was regulated by the proposed
MPPT technique, and at SP1, the output power reached 15,000 W, which was the MPP.
In the proposed MPPT technique, at t = 0.5 s, the SP suddenly changed from SP1 to SP2.
This resulted in locating the new MPP at 1265 V, where the array power output changed
Electronics 2023, 12, x FOR PEER REVIEW 15 of 22
to 8000 W, as shown in Figure 13 successfully. Furthermore, the planned MPPT technique
instantly controlled the duty ratio of the buck-boost converter.

Figure 13.Voltage
Figure13. Voltagetracking
trackingwithout
withoutuncertainties.
uncertainties.

Moreover,
Moreover,atatthe
thesame
sametime
timefor
forthe
thesame system,
same system,thethe
simulations runrun
simulations forfor
conventional
conventional
techniques (P&O and PID). As shown in Figure 13, the conventional
techniques (P&O and PID). As shown in Figure 13, the conventional techniques techniques failed to to
failed
track instantly the voltage at the time of dynamic change in the shading pattern; that was at
track instantly the voltage at the time of dynamic change in the shading pattern; that was
0.5 s when SP changed from SP1 to SP2. It can be noted that in this scenario the simulations
at 0.5 s when SP changed from SP1 to SP2. It can be noted that in this scenario the simula‐
were carried out without adding uncertainties in the system and it is clear from the results
tions were carried out without adding uncertainties in the system and it is clear from the
that the proposed method obtained success in tracking the MPP with a rapid response
results that the proposed method obtained success in tracking the MPP with a rapid re‐
under different shading patterns.
sponse under different shading patterns.
8.2. Performance Analysis with Uncertainities
8.2. Performance
The SimulinkAnalysis with Uncertainities
model which is described already is used by adding some uncertainties
Thesystem
into the Simulink model
to check thewhich is described
robustness already
and to make is used by adding
a comparison some
of it with uncertainties
other existing
into the system
techniques shown to in
check the14.
Figure robustness and to
Uncertainties inmake a comparison
the shape of it with
of capacitance other existing
and inductance
were added shown
techniques to the parameters
in Figure 14.ofUncertainties
the buck-boost inconverter.
the shape A of capacitance
capacitance of uF was
58inductance
and
added in parallel
were added to the C2 and an inductance
withparameters of 30 mHconverter.
of the buck‐boost was addedAincapacitance
series with an 58 𝑢F was
ofinductor
which
addedwas conducted
in parallel withfor
C20.7 to an
and 0.8 inductance
s by mean ofofsignal
30 mH builder and switches.
was added in series with an induc‐
tor which was conducted for 0.7 to 0.8 s by mean of signal builder and switches.
 The Simulink model which is described already is used by adding some uncertainties
into the system to check the robustness and to make a comparison of it with other existing
techniques shown in Figure 14. Uncertainties in the shape of capacitance and inductance
were added to the parameters of the buck‐boost converter. A capacitance of 58 𝑢F was
Electronics 2023, 12, 1720 15 of 22
added in parallel with C2 and an inductance of 30 mH was added in series with an induc‐
tor which was conducted for 0.7 to 0.8 s by mean of signal builder and switches.

Electronics 2023, 12, x FOR PEER REVIEW

Figure14.
Figure 14.Power
Poweranalysis
analysiswithout
without uncertainties.
uncertainties. 16 of 22

The
Thevoltage
voltagetracking
trackingand andthe
thechanging
changing shading
shading pattern
patternfrom
fromSP1 to SP2
SP1 forfor
to SP2 thethe
pro-pro‐
posed
posed technique and
andwas other
other existing
existing methods
methods (P&O (P&O and PID) are displayed in 15
and 16. Initially, SP1 applied where the MPP wasand PID)at
located are displayed
1645 V and itsinsubsequent
Figures
Figures 15 and 16. Initially, SP1 was applied where the MPP was located at 1645 V and its
output power was 15,000 W. This proposed methodology successfully operated at this
subsequent output power was 15,000 W. This proposed methodology successfully operated
point. The P&O
at this point. Theworked
P&O workedat 1630atV 1630
and the calculated
V and value of
the calculated the output
value was 14,500
of the output was W.
The proposed
14,500 W. The proposed was able was
techniquetechnique to accurately locate thelocate
able to accurately new MPP
the new at 1265
MPPVatand 8000
1265 V W
despite the uncertainties caused by changes in the shading patterns from
and 8000 W despite the uncertainties caused by changes in the shading patterns from SP1 SP1 to SP2. How‐
ever,
to SP2.itHowever,
is importantit is to note thattotraditional
important methods failed
note that traditional methods to perform
failed to well under
perform wellthese
conditions,
under often oscillating
these conditions, around local
often oscillating peaks
around andpeaks
local experiencing significant
and experiencing chattering,
significant
as seen in Figure
chattering, as seen16. The proposed
in Figure technique,technique,
16. The proposed however, however,
was able wasto precisely identify the
able to precisely
optimalthe
identify operating
optimal voltage,
operating resulting
voltage, in a significant
resulting enhancement
in a significant in system
enhancement efficiency.
in system
Both the proposed
efficiency. and traditional
Both the proposed methods
and traditional are illustrated
methods in Figures
are illustrated 15 and
in Figures 15 16,
andrespec‐
16,
respectively,
tively, showingshowing the variations
the variations in PV interminal
PV terminal voltage,
voltage, output
output power,
power, andand
thethe duty
duty cycle
cycle
duringduring the transition
the transition fromfrom
SP1 SP1 to SP2.
to SP2.

Figure15.
Figure 15.Voltage
Voltagetracking
tracking with
with uncertainties.
uncertainties.
Electronics 2023, 12, 1720 16 of 22

Figure 15. Voltage tracking with uncertainties.

Figure16.
Figure 16.Power
Poweranalysis
analysiswith
withuncertainties.
uncertainties.
Electronics 2023, 12, x FOR PEER REVIEW 17 of 22

The
Theresults
resultsindicate
indicateclearly
clearlythat
thatthetheproposed
proposedtechnique
techniqueshowed
showedminimum
minimum chattering
chattering
as
as compared
comparedtotoP&O
P&Oand
andPID
PIDforfor0.7
0.7toto0.8 s when
0.8 s whenuncertainties were
uncertainties added;
were it could
added; bearbear
it could
abrupt
8.3. changes
Duty Cycle which
and demonstrated
Fault Analysis control robustness.
abrupt changes which demonstrated control robustness.
Figure
8.3. Duty 17and
Cycle demonstrates that the proposed technique was successful in the instanta‐
Fault Analysis
neous control of the duty cycle. When the shading pattern changes from SP1 to SP2 at 0.5 s
Figure 17 demonstrates that the proposed technique was successful in the instanta-
then the duty cycle changes from 0.21 to 0.35. On the other hand, the error convergence
neous control of the duty cycle. When the shading pattern changes from SP1 to SP2 at 0.5 s
then the dutywhen
was tested cyclethe shading
changes frompatterns changed
0.21 to 0.35. from
On the SP1
other to SP2,
hand, the which is shown inwas
error convergence Figure
18. when the shading patterns changed from SP1 to SP2, which is shown in Figure 18.
tested

Figure17.
Figure 17.Duty
Dutycycle
cycleatatSP1
SP1and
and SP2.
SP2.
Electronics 2023, 12, 1720 17 of 22

Figure 17. Duty cycle at SP1 and SP2.

Figure 18. Error in SP1 and SP2.

Figure 18. Error in SP1 and SP2.
9.
9. Comparative
ComparativeAnalysis
Analysisof of
Proposed Technique
Proposed Technique
The
Theproposed
proposedsystem
systemhas been
has validated
been through
validated a practical
through experimental
a practical setup.setup.
experimental The The
PV emulator used in this study is based on the single diode model, where the
PV emulator used in this study is based on the single diode model, where the PV system PV system
acts
acts as
as aa current
currentsource
sourcewith anan
with antiparallel
antiparalleldiode andand
diode intrinsic resistances.
intrinsic Literature
resistances. Literature
shows that a PV cell is essentially a voltage source dependent on current, and its output
shows that a PV cell is essentially a voltage source dependent on current, and its output
current and open‐circuit voltage (VOC) vary with changes in irradiance and temperature.
current and open-circuit voltage (VOC ) vary with changes in irradiance and temperature.
Therefore, a DC power source with high current and low sensitivity can emulate the elec‐
Therefore, a DC power source with high current and low sensitivity can emulate the
trical characteristics of a PV system. The superposition of constant current, constant volt‐
electrical characteristics of a PV system. The superposition of constant current, constant
age, and diode activation effectively mimics the electrical behavior of a PV cell. The load
voltage,
was changed
Electronics 2023, 12, x FOR PEER REVIEW
andusing
diodea 500
activation effectively
W variable resistor, mimics
which inthe electrical
practical termsbehavior
alters theof a PV
maximum cell. The
18 of 22
load was changed using a 500 W variable resistor, which in practical
power delivery to the load and forces the algorithm to maximize the power at a changingterms alters the
maximum power delivery to the load and forces the algorithm to maximize the power at a
changing load following the maximum power delivery theorem. The impact of irradiance
load following the maximum power delivery theorem. The impact of irradiance was eval‐
was evaluated by suddenly changing the voltage of the DC current source. Figure 19
uated by suddenly changing the voltage of the DC current source. Figure 19 illustrates the
illustrates the physical connections among the PV system components such as the DC—DC
physical connections among the PV system components such as the DC—DC buck‐boost
buck-boost converter, sensors, microcontrollers for MPPT control, data acquisition, and
converter, sensors, microcontrollers for MPPT control, data acquisition, and load. The per‐
load. The of
formance performance
the proposed of MLNFN
the proposed
TSMC MLNFN TSMC was
was compared compared
to that to that of
of a ZN‐tuned PIDa ZN-tuned
con‐
troller. The emulation function was limited by the variable current generation for the PV for
PID controller. The emulation function was limited by the variable current generation
the PV module.
module. Table 7 Table 7 lists
lists the the of
values values of the components
the components used forused for the practical
the practical application
application of
of control. Figure 20 shows the layout of the experimental
control. Figure 20 shows the layout of the experimental setup. setup.

Figure
Figure 19.
19.An
Anexperimental
experimentalsetup with
setup a low‐cost
with PV PV
a low-cost emulator.
emulator.
 troller. The emulation function was limited by the variable current generation for the PV
module. Table 7 lists the values of the components used for the practical application of
control. Figure 20 shows the layout of the experimental setup.

Electronics 2023, 12, 1720 18 of 22

Table 7. Specifications of hardware components.

Parameters Symbols
Load, (RL ) 5,10 Ω, 300 W
DC source 1 PS305
DC source 2 MS305-D dual channel
Switching frequency, ( f ) 61 kHz
Inductor (L) 1 mH
Output capacitor (Cin ) 1000 µF
Input capacitor (Cout ) 100 µF
Oscilloscope Tektronix TDS-3052B
Power diode PHY 10SQ04
Voltage sensor B25 voltage sensor
Current sensor module ACS172
Micro-controller ATmega 2560/328
MOSFET IRF730
Figure 19. An experimental setup with a low‐cost PV emulator.

Figure20.
Figure 20.Experimental
Experimentalsetup
setupforfor MPPT.
MPPT.

Table 7. Specifications
Figure of hardware
21 shows tracked powercomponents.
with dynamic load conditions by MLNFN TSMC and
PID controllers. PID takes up to 250 ms for final GMPP settling as shown in Figure 21B.
Parameters Symbols
Oscillations which were unavoidable were noticed after 250 ms represented in magnification
window, in addition to Load, (RL)maximum power was achieved with
this less 5,10
PID.Ω,It300 W to power
added
DC source 1 PS305
loss and decreased inefficiency. MLNFN TSMC tracks GM faster as compared to PID and
DC source
settles at GMPP in 100–120 2 shown in Figure 21A. MLNFN
ms as MS305‐D
TSMCdual showedchannel
minimum
oscillations after GMPP detection
Switching frequency, and consequently generated the least61power
(𝑓) kHz loss in steady
and dynamic states Inductor
showing an (L)8.2% higher average power and 60% less 1 mHcontrol time. The
performance inOutput
transition and final
capacitor (𝐶 ) steady states was observed in the experiment,
1000 μF similar
to the anticipated behavior in the mathematical model. Negligible fluctuations and the
Input capacitor (𝐶 ) 100 μF
least settling time in experimental results reiterated better performance of MLNFN TSMC.
Figure 22A,B, shows the reference voltages computed by the MLNFN TSMC and PID
controller, respectively.
 minimum oscillations after GMPP detection and consequently generated the least power
loss in steady and dynamic states showing an 8.2% higher average power and 60% less
control time. The performance in transition and final steady states was observed in the
experiment, similar to the anticipated behavior in the mathematical model. Negligible
fluctuations and the least settling time in experimental results reiterated better perfor‐
Electronics 2023, 12, 1720 19 of 22
mance of MLNFN TSMC. Figure 22A,B, shows the reference voltages computed by the
MLNFN TSMC and PID controller, respectively.

(A)

Electronics 2023, 12, x FOR PEER REVIEW (B) 20 of 22

Figure
Figure 21. Experimental
21. Experimental power
power transient
transient of PV
of PV System
System (A)(A) MLNFN
MLNFN TSMC
TSMC (B) (B) PID.
PID.

(A)

Figure 22. Cont.

Electronics 2023, 12, 1720 20 of 22

(A)

(B)

FigureFigure 22. Experiment

22. Experiment voltage
voltage transient
transient of System
of PV PV System
(A)(A) MLNFN
MLNFN TSMC
TSMC (B)(B) PID.
PID.

10. Common
10. Common Performance
Performance Analysis
Analysis andand Discussions
Discussions
This study
This study delvesdelves
into into the shortcomings
the shortcomings of conventional
of conventional PID
PID control‐basedmethods
control-based methods
for tracking
for tracking the MPP the in
MPPPVin PV systems
systems during during
PSCs.PSCs.
TheseThese techniques
techniques are unable
are unable to circum‐
to circumvent
vent local(LM)
local maxima maxima
traps(LM)
andtraps
oftenandleadoften lead tofluctuations
to voltage voltage fluctuations that impede
that impede the inte‐
the integration
gration of large‐scale
of large-scale PV systems into power grids. To address these challenges, in recent recent
PV systems into power grids. To address these challenges, in years
years the researchers
the researchers have proposed have proposed a new technique
a new technique called called the intelligent
the intelligent modified
modified shuf‐
shuffled
fled frog leaping algorithm (IMSFLA) which has been found to be more
frog leaping algorithm (IMSFLA) which has been found to be more effective in tracking the effective in track‐
GMPP ing the an
with GMPP with anofefficiency
efficiency 99%. IMSFLA of 99%.also
IMSFLA also demonstrates
demonstrates minimalminimal oscillations
oscillations at the
at the GMPP resulting in increased power output to the load. The
GMPP resulting in increased power output to the load. The IMSLFA algorithm operates IMSLFA algorithm op‐
erates by dividing the problem into several sub‐problems, which are then optimized in
by dividing the problem into several sub-problems, which are then optimized in parallel.
parallel. The algorithm uses a combination of local and global search strategies to explore
The algorithm uses a combination of local and global search strategies to explore the search
the search space efficiently. The local search strategy helps the algorithm to converge
space efficiently. The local search strategy helps the algorithm to converge quickly, while
quickly, while the global search strategy helps it to avoid local optima and find the global
the global search strategy helps it to avoid local optima and find the global MPP. Compared
MPP. Compared with conventional techniques, the reported IMSFLA provides improved
with conventional techniques, the reported IMSFLA provides improved accuracy, faster
accuracy, faster tracking speed, and robustness to noise.
tracking speed, and robustness
Our proposed methodto of noise.
MLNFN TSMC has been verified as robust in dynamic as
Our proposed method of MLNFN
well as static operating conditions through TSMC has been verified
hardware as robust
experiments. in dynamic
These experiments as
well as static operating conditions through hardware experiments.
show that the proposed technique performs well under real‐world conditions and loadThese experiments
showvariations,
that the proposed
resulting technique
in 56.1% less performs well
variability andunder
only areal-world conditions
2–3 W standard and at
deviation load
the
variations,
MPP andresulting
found in 56.1% the
to mimic lessIMSFLA
variability and only a 2–3 W standard deviation at the
performance.
MPP and found to mimic the IMSFLA performance.
The purposed technique combines the advantages of fuzzy logic control, neural
network-based control, and terminal sliding mode control to achieve optimal performance.
In real-world partial shading conditions and load variation, the performance of the neuro-
fuzzy terminal sliding mode MPPT technique depends on several factors, such as the
complexity of the system, the quality of the sensors, and the accuracy of the modeling of the
PV system. However, the technique has been shown to be effective in improving the perfor-
mance of PV systems under partial shading conditions and load variation. The purposed
technique improves the efficiency of the PV system under partial shading conditions and
load variation, compared to other MPPT techniques. The technique was able to adjust the
duty cycle of the DC—DC converter to ensure that the PV system operated at the MPP, even
under partial shading conditions. The results also showed that the proposed technique
was able to maintain a stable output voltage and current under partial shading conditions
and load variation. The neuro-fuzzy logic was able to adjust the controller parameters in
real-time based on the input signals, which ensured the stability of the system. Furthermore,
the terminal sliding mode control was able to provide fast and accurate tracking of the
MPP, even under dynamic partial shading conditions.
Electronics 2023, 12, 1720 21 of 22

11. Conclusions
This study proposes a two-stage MPPT approach for PV systems operating under PSCs.
A NFN is employed to enhance the reference voltage generation for MPPT. The proposed
strategy guarantees finite-time convergence of MPP voltage tracking and resolves the issue
of multiple peaks caused by shading conditions, in contrast to traditional TSMC methods.
The method is tested through real-time experiments and numerical simulations and was
found to significantly improve MPPT performance, even under rapidly changing irradiance
and temperature conditions. The proposed robust TSMC approach is also more resilient, as
demonstrated by simulations under varying weather conditions and uncertainties. The
results show that the proposed method can quickly and accurately track the MPP, even in
the presence of uncertainties, when compared to P&O, PID and incremental conductance
conventional controllers.

Author Contributions: Data curation, A.; formal analysis, M.B.Q. and W.A.; funding acquisition,
M.A.B.F.; investigation, A., W.A. and M.M.K.; methodology, A. and M.B.Q.; Project administration,
M.B.Q. and M.A.B.F.; Software, A. and W.A.; supervision, M.B.Q. and R.N.; validation, M.M.K.;
visualization, R.N.; writing—original draft, A. and M.B.Q.; writing—review and editing, M.M.K. and
M.A.B.F. All authors have read and agreed to the published version of the manuscript.
Funding: This research received no external funding.
Data Availability Statement: Not applicable.
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Sampaio, P.G.V.; González, M.O.A. Photovoltaic solar energy: Conceptual framework. Renew. Sustain. Energy Rev. 2017, 74,
590–601. [CrossRef]
2. Kittner, N.; Lill, F.; Kammen, D.M. Energy storage deployment and innovation for the clean energy transition. Nat. Energy 2017,
2, 17125. [CrossRef]
3. Verma, D.; Nema, S.; Shandilya, A.M.; Dash, S.K. Maximum power point tracking (MPPT) techniques: Recapitulation in solar
photovoltaic systems. Renew. Sustain. Energy Rev. 2016, 54, 1018–1034. [CrossRef]
4. Rezk, H.; Fathy, A.; Abdelaziz, A.Y. A comparison of different global MPPT techniques based on meta-heuristic algorithms for
photovoltaic system subjected to partial shading conditions. Renew. Sustain. Energy Rev. 2017, 74, 377–386. [CrossRef]
5. Liu, J.; Li, J.; Wu, J.; Zhou, W. Global MPPT algorithm with coordinated control of PSO and INC for rooftop PV array. J. Eng. 2017,
13, 778–782. [CrossRef]
6. Yeung, R.S.C.; Chung, H.S.H.; Tse, N.C.F.; Chuang, S.T.H. A global MPPT algorithm for existing PV system mitigating suboptimal
operating conditions. Sol. Energy 2017, 141, 145–158. [CrossRef]
7. Hu, Y.; Cao, W.; Wu, J.; Ji, B.; Holliday, D. Thermography-based virtual MPPT scheme for improving PV energy efficiency under
partial shading conditions. IEEE Trans. Power Electron. 2014, 11, 5667–5672. [CrossRef]
8. Mohanty, S.; Subudhi, B.; Ray, P.K. A new MPPT design using grey wolf optimization technique for photovoltaic system under
partial shading conditions. IEEE Trans. Sustain. Energy 2015, 7, 181–188. [CrossRef]
9. Tamir, S. Particle Swarm Optimization with Targeted Position-Mutated Elitism (PSO-TPME) for Partially Shaded PV Systems.
Sustainability 2023, 15, 3993.
10. Loukriz, A.; Haddadi, M.; Messalti, S. Simulation and experimental design of a new advanced variable step size Incremental
Conductance MPPT algorithm for PV systems. ISA Trans. 2016, 62, 30–38. [CrossRef]
11. Safari, A.; Mekhilef, S. Incremental Conductance MPPT Method for PV Systems. In Proceedings of the 2011 24th Canadian
Conference on Electrical and Computer Engineering (CCECE), Niagara Falls, ON, Canada, 8–11 May 2011.
12. Saravanan, S.; Babu, N.R. Maximum power point tracking algorithms for photovoltaic system system—A review. Renew Sustain.
Energy Rev. 2016, 57, 192–204. [CrossRef]
13. Mao, M.; Cui, L.; Zhang, Q.; Guo, K.; Zhou, L.; Huang, H. Classification and summarization of solar photovoltaic MPPT
techniques: A review based on traditional and intelligent control strategies. Energy Rep. 2020, 6, 1312–1327. [CrossRef]
14. Abadi, I.; Imron, C.; Noriyati, R.D. Noriyati. Implementation of Maximum Power Point Tracking (MPPT) Technique on Solar
Tracking System Based on Adaptive Neuro-Fuzzy Inference System (ANFIS). EDP Sci. 2018, 43, 01014.
15. Armghan, H.; Ahmad, I.; Armghan, A.; Khan, S.; Arsalan, M. Backstepping based non-linear control for maximum power point
tracking in photovoltaic system. Sol. Energy 2018, 159, 134–141.
16. Katche, M.L.; Makokha, A.B.; Zachary, S.O.; Adaramola, M.S. A Comprehensive Review of Maximum Power Point Tracking
(MPPT) Techniques Used in Solar PV Systems. Energies 2023, 16, 2206. [CrossRef]
Electronics 2023, 12, 1720 22 of 22

17. Craciunescu, D.; Fara, L. Investigation of the Partial Shading Effect of Photovoltaic Panels and Optimization of Their Performance
Based on High-Efficiency FLC Algorithm. Energies 2023, 16, 1169. [CrossRef]
18. Abo-Khalil, A.G.; El-Sharkawy, I.I.; Radwan, A.; Memon, S. Influence of a Hybrid MPPT Technique, SA-P&O, on PV System
Performance under Partial Shading Conditions. Energies 2023, 16, 577.
19. Farayola, A.M.; Sun, Y.; Ali, A. Global maximum power point tracking and cell parameter extraction in Photovoltaic systems
using improved firefly algorithm. Energy Rep. 2022, 8, 162–186. [CrossRef]
20. Gong, L.; Hou, G.; Huang, C. A two-stage MPPT controller for PV system based on the improved artificial bee colony and
simultaneous heat transfer search algorithm. ISA Trans. 2023, 132, 428–443. [CrossRef]
21. Hassan, A.; Bass, O.; Masoum, M.A. An improved genetic algorithm based fractional open circuit voltage MPPT for solar PV
systems. Energy Rep. 2023, 9, 1535–1548. [CrossRef]
22. Punitha, K.; Devaraj, D.; Sakthivel, S. Artificial neural network based modified incremental conductance algorithm for maximum
power point tracking in photovoltaic system under partial shading conditions. Energy 2013, 62, 330–340. [CrossRef]
23. Hiren, P.; Agarwal, V. Maximum power point tracking scheme for PV systems operating under partially shaded conditions. IEEE
Trans. Ind. Electron. 2008, 55, 1689–1698.
24. Koad, R.B.; Zobaa, A.F.; El-Shahat, A. A novel MPPT algorithm based on particle swarm optimization for photovoltaic systems.
IEEE Trans. Sustain. Energy 2016, 8, 468–476. [CrossRef]
25. El-Helw, H.M.; Magdy, A.; Marei, M.I. A hybrid maximum power point tracking technique for partially shaded photovoltaic
arrays. IEEE Access 2017, 5, 11900–11908. [CrossRef]
26. Sun, Y.; Peng, Y.; Deng, F. Improved SPSO-based Parameter Identification of Solar PV Cells IV Model. In Proceedings of the 2017
International Conference on Computer Systems, Electronics and Control (ICCSEC), Dalian, China, 25–27 December 2017; IEEE:
Piscataway, NJ, USA, 2017; pp. 1738–1742.
27. Nazri, N.S.; Fudholi, A.; Ruslan, M.H.; Sopian, K. Mathematical modeling of photovoltaic thermal-thermoelectric (PVT-TE) air
collector. Int. J. Power Electron. Drive Syst. 2018, 9, 795.
28. Bressan, M.; Gutierrez, A.; Gutierrez, L.G.; Alonso, C. Development of a real-time hot-spot prevention using an emulator of
partially shaded PV systems. Renew. Energy 2018, 127, 334–343. [CrossRef]
29. Bingöl, O.; Özkaya, B. Analysis and comparison of different PV array configurations under partial shading conditions. Sol. Energy
2018, 160, 336–343. [CrossRef]
30. Ahsan, S.; Niazi, K.A.K.; Khan, H.A.; Yang, Y. Hotspots and performance evaluation of crystalline-silicon and thin-film photo-
voltaic modules. Microelectron. Reliab. 2018, 88, 1014–1018. [CrossRef]
31. Rajurkar, S.; Verma, N.K. Developing deep fuzzy network with Takagi Sugeno fuzzy inference system. In Proceedings of the 2017
IEEE International Conference on Fuzzy Systems (FUZZ-IEEE), Naples, Italy, 9–12 July 2017; IEEE: Piscataway, NJ, USA, 2017;
pp. 1–6.
32. Arsalan, M.; Iftikhar, R.; Ahmad, I.; Hasan, A.; Sabahat, K.; Javeria, A. MPPT for photovoltaic system using nonlinear backstepping
controller with integral action. Sol. Energy 2018, 170, 192–200. [CrossRef]
33. Kaouane, M.; Boukhelifa, A.; Cheriti, A. Regulated output voltage double switch Buck-Boost converter for photovoltaic energy
application. Int. J. Hydrog. Energy 2016, 41, 20847–20857. [CrossRef]

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual
author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to
people or property resulting from any ideas, methods, instructions or products referred to in the content.