electronics

Article
An Artificial Neural Network for Solar Energy Prediction and
Control Using Jaya-SMC
Mokhtar Jlidi 1 , Faiçal Hamidi 1, *, Oscar Barambones 2, * , Rabeh Abbassi 3 , Houssem Jerbi 4 ,
Mohamed Aoun 1 and Ali Karami-Mollaee 5

1 Laboratory Modélisation, Analyse et Commande des Systèmes, University of Gabes, Gabes LR16ES22, Tunisia
2 Automatic Control and System Engineering Department, University of the Basque Country, UPV/EHU,
Nieves Cano 12, 01006 Vitoria-Gasteiz, Spain
3 Department of Electrical Engineering, College of Engineering, University of Ha’il, Hail 1234, Saudi Arabia
4 Department of Industrial Engineering, College of Engineering, University of Ha’il, Hail 1234, Saudi Arabia
5 Faculty of Electrical and Computer Engineering, Hakim Sabzevari University, Sabzevar 96186-76115, Iran
* Correspondence: [email protected] (F.H.); [email protected] (O.B.); Tel.: +34-94-501-3235 (O.B.);
Fax: +34-94-501-3270 (O.B.)

Abstract: In recent years, researchers have focused on improving the efficiency of photovoltaic
systems, as they have an extremely low efficiency compared to fossil fuels. An obvious issue
associated with photovoltaic systems (PVS) is the interruption of power generation caused by changes
in solar radiation and temperature. As a means of improving the energy efficiency performance of
such a system, it is necessary to predict the meteorological conditions that affect PV modules. As part
of the proposed research, artificial neural networks (ANNs) will be used for the purpose of predicting
the PV system’s current and voltage by predicting the PV system’s operating temperature and
radiation, as well as using JAYA-SMC hybrid control in the search for the MPP and duty cycle single-
ended primary-inductor converter (SEPIC) that supplies a DC motor. Data sets of size 60538 were
used to predict temperature and solar radiation. The data set had been collected from the Department
of Systems Engineering and Automation at the Vitoria School of Engineering of the University of the
Citation: Jlidi, M.; Hamidi, F.;
Basque Country. Analyses and numerical simulations showed that the technique was highly effective.
Barambones, O.; Abbassi, R.; Jerbi,
In combination with JAYA-SMC hybrid control, the proposed method enabled an accurate estimation
H.; Aoun, M.; Karami-Mollaee, A. An
of maximum power and robustness with reasonable generality and accuracy (regression (R) = 0.971,
Artificial Neural Network for Solar
Energy Prediction and Control Using
mean squared error (MSE) = 0.003). Consequently, this study provides support for energy monitoring
Jaya-SMC. Electronics 2023, 12, 592. and control.
https://doi.org/10.3390/
electronics12030592 Keywords: JAYA algorithm; forecasting; artificial neural networks; sliding mode control; PEMFC;
MPPT; SEPIC chopper
Academic Editors: Luigi Scarcello,
Carlo Mastroianni and
Teemu Leppänen

Received: 22 December 2022 1. Introduction

Revised: 18 January 2023
Nowadays, the prediction of the developments that will occur in all areas is becoming
Accepted: 21 January 2023
a critical necessity. Therefore, the techniques that have merged to predict a future result or
Published: 25 January 2023
ensure the so-called forecasting are increasingly being demanded. Among such techniques,
the artificial neural network (ANN) has been used as a forecasting tool in many applications.
Indeed, it has been used in the tourism sector to forecast the number of tourists or hotel
Copyright: © 2023 by the authors.
stays in a country [1]. It has also been used in the financial trading [2] and energy sector,
Licensee MDPI, Basel, Switzerland. including power usage and especially renewable energy [3–5]. The literature has shown
This article is an open access article many published studies highlighting the performance of neural network-based methods
distributed under the terms and compared to other forecasting methods, particularly with high-frequency data. The present
conditions of the Creative Commons investigation focuses onrenewable energies application because it is among the most
Attribution (CC BY) license (https:// important fields attracting the researchers worldwide. This is justified by the fact that such
creativecommons.org/licenses/by/ energies are sustainable and preserving the ecosystem. Solar energy and wind energy are
4.0/). the most important and widely used renewable energy sources. Solar energy resources are

Electronics 2023, 12, 592. https://doi.org/10.3390/electronics12030592 https://www.mdpi.com/journal/electronics

Electronics 2023, 12, 592 2 of 26

considered an efficient source of sustainable energy. Photovoltaic solar energy is converted

into electricity using photovoltaic cells. Today, the energy extracted from photovoltaic
(PV) power plants can be off-grid or integrated into the electrical grid [6], especially
when they are connected to other energy sources to form a hybrid system [7,8]. As an
innovative forecasting framework, FE-SVR-mFFO has been proposed as a hybrid technique
for improving strategic decision making in smart grid applications [9]. Based on the FE-
SVR-mFFO results, it has been demonstrated that it is effective in terms of stability, accuracy,
and convergence rate.
To achieve the sizing of the photovoltaic generator, its produced energy under con-
tinuous weather changes should be estimated. Consequently, weather changes should be
predicted in order to facilitate the PV system control and monitoring. There are two types
of forecasting methods: the qualitative methods based on sensory emotions, doubts, experi-
ences, and opinions; and the quantitative methods depending entirely on mathematical
calculations. In relation to the studied application of PV systems, the second method is the
most commonly used method to predict the temperature and the solar irradiation.
Despite the advantages of solar energy, photovoltaic generators have some drawbacks,
such as a lower efficiency compared to fossil energy sources, as well asan instability of the
produced energy since it is influenced by the changes in the atmospheric conditions. This
issue seriously affects the grid stability. The energy produced by a photovoltaic generator
changes with sunlight and temperature and can also provide a better energy conversion
at a specified operating point. This point is called the maximum power point (MPP). The
researchers designed several techniques for tracking the MPP to achieve voltage and current
regulation [10]. It has been mentioned in [11,12] that the maximum power point tracking
(MPPT) control techniques based on electric current measurement are used; however, they
have some drawbacks related to employed analogue controllers. However, digital MPPT
controllers are designed by algorithms, as investigated in [7,11,13,14]. In the literature,
many methods to search the MPP point have been proposed. The most famous one is the
perturb and observe (P&O) algorithm, which is the oldest and simplest one. In [15], it
has been indicated that the algorithm (P&O) is generally dependent on initial conditions
and exhibits fluctuations throughout the optimum point. Otherwise, this method has
a bad performance in the case of sudden changes in meteorological conditions [16–20].
Taking that into consideration, another MPPT technique called incremental conductance
(IC) has been proposed to overcome the limits of the P&O algorithm. The IC is considered
more complex than the P&O [21]. The algorithms based on the measurement of the open
circuit voltage (Voc) or the short circuit current (Isc) are veritably simple and easy to apply.
Nevertheless, their main problem is the power loss and the transmission interruption
during Voc and Isc measurement [22,23].
To overcome this problem, an experimental cell of the same nature as the PV panel
cells is used. In addition, locating the optimal operating point is particularlychallenging.
Therefore, these methods are only approximations that do not give enough precision, and
therefore, the system does not necessarily operate at the optimal point. In this context,
the technique based on fuzzy logic control (FLC) theory is important and effective [24].
Indeed, this technique works at the optimal point without oscillations [25]. However, the
implementation of the FLC technique is more complex than classical algorithms and its
efficiency depends largely on the rule table.
In a more developed context, many optimization techniques, some based on the
heuristic approach and others on the meta-heuristic approach, have been used for the
implementation of MPPT techniques. Swarm-based algorithms (particle swarm, artificial
bee colony,...), and trajectory-based algorithms (Tabu search, hill climbing,...), as well as
evolutionary algorithms (genetic algorithm,...) have been suggested [26,27]. In particular,
the evolutionary intelligence and swarm based algorithms are probabilistic algorithms
that require common control parameters and are very efficient for MPP point tracking.
Therefore, the choice of a reliable and consistent MPPT technique was targeted towards the
JAYA algorithm [28]. This algorithm was formulated by RAO in 2016 to solve constrained
Electronics 2023, 12, 592 3 of 26

and unconstrained optimization problems, based on the concept that the solution obtained
for a given problem should move towards the best solution and should avoid the worst
solution [29]. The JAYA algorithm is a definitive, parameter-free algorithm.
In contrast to other meta-heuristic algorithms, the JAYA algorithm does not require
any special parameters. Compared to, for instance, GA, PSO, CPSO, and GWO, each
algorithm requires careful selection of parameters. It is because they affect the accuracy of
the search for the optimal value. A further advantage of JAYA is that it has a very simple
structure and has been shown to be effective at solving optimization problems [30,31].
Since SMC has a simple algorithm and a high degree of robustness, it has been widely
used for nonlinear control systems [32]. As a result of the simplicity of both methods, as
well as their effectiveness, JAYA and SMC have been combined in this work. It has been
shown in [33] that gradient optimization techniques are combined with the PI controller
to determine the optimal point and control for SEPIC. However, due to the complexity of
this controller, it is difficult to extract the PI gains. Furthermore, the gradient optimization
method requires special parameters to reduce error, which are difficult to determine. In
this way, hybridization between JAYA and SMC provides a simple and robust means for
controlling and determining the optimal point.
In this paper, the JAYA algorithm has been exploited to control the DC-DC converter
and ensure MPP point tracking. To control the duty cycle, there are many control methods.
Among the best-known methods are those that use PID or PI controllers [33,34]. These
methods are simple and easy to use as their parameters (P proportional, D derivative,
P integral) are relatively easy to measure. However, these methods are not practical in
industrial or energy environments as they have many non-constant inputs that change
continuously with the weather and external factors. As a result, these methods provide
limited performance in the presence of external disturbances and uncertainties. On the
other hand, sliding mode control (SMC) is an important technology in many dynamic
and complex domains, mainly because of its significant ability to reject uncertainties and
exclude external disturbances. SMC has been widely used for non-linear control systems
due to its simple algorithm and high robustness.
By predicting the operating temperature and radiation of the PVS, the proposed
research makes use of artificial neural network (ANN) technology to predict the current
and voltage. Further, the JAYA-SMC hybrid control system is designed to determine the
MPP and duty cycle of the single-ended primary-inductor converter (SEPIC) that supplies
the DC motor.
Even with the constant evolution of photovoltaic panel technology in the field of elec-
tricity generation, the storage of electrical energy for later use or to reduce the fluctuation of
solar production remains a great challenge for users of this technology when not supplied
by the grid. Possible solutions include batteries for storage, which are expensive and
difficult to maintain safely and need to be replaced frequently. To this end, other new and
more viable technologies for energy storage, such as systems that rely on hydrogen storage,
have become available due to their significant advantages over battery systems [35–37].
The underlying technology that makes this possible is hydrogen electrolysis, which is the
process where the use of low voltages is used to create reactions in various solutions. This
is the electrolytic process, which can be used to split a molecule of water into hydrogen
and oxygen [38].
Energy is of the utmost importance, as mentioned above, as such it is imperative to
develop management methods, the most notable of which are forecasting techniques and
MPPT detection. Recent years have seen the development of these methods. In this respect,
choosing the most appropriate MPPT method and making a prediction can be a challenge,
since each method has its own advantages and disadvantages. In light of these factors, it is
imperative to select simple, effective, and straightforward methods, particularly when there
are unpredictable weather conditions that can prevent some methods from performing
properly, e.g., GA and PSO. For control and monitoring, critical evaluation and analysis
are key. A specific idea presented in this paper is combining robust control and simple
 and MPPT detection. Recent years have seen the development of these methods. In this
respect, choosing the most appropriate MPPT method and making a prediction can be a
challenge, since each method has its own advantages and disadvantages. In light of
these factors, it is imperative to select simple, effective, and straightforward methods,
particularly when there are unpredictable weather conditions that can prevent some
Electronics 2023, 12, 592 4 of 26
methods from performing properly, e.g., GA and PSO. For control and monitoring, criti-
cal evaluation and analysis are key. A specific idea presented in this paper is combining
robust control and simple optimization techniques for energy prediction and control,
optimization
while techniques
maintaining for energy
robustness prediction and control, while maintaining robustness
and accuracy.
and accuracy.
This paper makes the following major contributions:
This•paper makes the following
Implementation major
of artificial contributions:
neural networks (ANN) to predict temperature
• and solarofradiation
Implementation artificial as it is one
neural of the (ANN)
networks most effective and
to predict efficient methods
temperature in
and solar
allasfields.
radiation it is one of the most effective and efficient methods in all fields.
• • Implementation
Implementation of JAYA-SMC based approach
of JAYA-SMC based to control DC-DC
approach converters
to control DC-DCaccording
convert-
to the maximum
ers according powerto point tracking concept
the maximum (MPPT).
power point tracking concept (MPPT).
The remainder
The remainderofofthethe paper
paper is organized
is organized as follows:
as follows: In theIn the second
second section, section, the
the method-
methodology is discussed,
ology is discussed, including including PV panelArtificial
PV panel modeling, modeling, Artificial
Neural NetworkNeural Network
modeling, and
modeling, and the implementation
the implementation of JAYA-SMC hybrid of JAYA-SMC
controllershybrid
for MPPcontrollers
extractionfor MPP
and extraction
SEPIC control.
and SEPIC control. Detailed simulation results and discussions are presented
Detailed simulation results and discussions are presented in Section 3. This paper ends in Section
III.
with This paper ends
concluding with concluding
statements statements andfor
and recommendations recommendations
future work. for future work.

2. Methodology
2. Methodology
The topology
The topology of the studied system is depicted in Figure 1.

Figure
Figure 1.
1. Topology
Topology of
of the
the studied
studied system.
system.

2.1. PV
2.1. PVPanel
PanelModeling
Modeling
The solar cell is
The solar cell is aa PN
PN junction
junction semiconductor.
semiconductor. WhenWhen exposed
exposed to to light,
light, it
it generates
generates aa
direct electric
direct electric current.
current. The The generated
generated current
current varies
varies slightly
slightly linearly
linearly with
with solar
solar radiation
radiation
variation. The commonly used circuit model of a PV cell consists of a
variation. The commonly used circuit model of a PV cell consists of a current source thatcurrent source that
Electronics 2023, 12, 592 5 of 28
depends on the values of solar radiation and temperature in parallel with a diode and aa
depends on the values of solar radiation and temperature in parallel with a diode and
shunt resistor
shunt resistor (R
(Rshsh),),that
thatare
areininseries
series with
with aa second
second resistor
resistor (R
(Rss))(Figure
(Figure2).
2).

Figure 2. PV panel equivalent circuit.

According
AccordingtotoKirchhoff’s current
Kirchhoff’s law, the
current law,generated current I is
the generated represented
currentI by Equation by
is represented (1)
Equation (1) q(V + IRs ) V + IRs
I = I ph − I0 (e aKTN s − 1) −
q (V + IRs )
(1)
V R+shIR
aKTN s
I = I ph − I 0 (e − 1) − s
(1)
Rsh
Where V is the voltage at the PV generator terminals, q is the charge of an electron
(1.602.10-19 C), a is the PV cells ideality factor, Ns is the number of PV cells connected in
series, K is the Boltzmann’s constant (1.38.10-23e-23J/K), NS is the number of series cells,
Electronics 2023, 12, 592 5 of 26

where V is the voltage at the PV generator terminals, q is the charge of an electron

(1.602 × 10−19 C), a is the PV cells ideality factor, NS is the number of PV cells connected in
series, K is the Boltzmann’s constant (1.38 × 10−23 J/K), NS is the number of series cells,
Iph is the photocurrent, described by Equation (2), and I0 is the saturation current of the
diode expressed by (3).
I + Ki ( T − TSTC )
I ph = G SC (2)
GSTC
where GSTC and TSTC are, respectively, the irradiation and temperature under standard
test conditions, G and T are their reel values, ISC is the short-circuit current, and Ki is the
temperature coefficient for short-circuit current.

ISC + Ki ( T − TSTC )
I0 = q[VOC +Kv ( T − TSTC )]
(3)
e aKTNS
−1
where Voc is the open-circuit voltage and KV is the temperature coefficient for open cir-
cuit voltage.
As can be seen in Equation (2), the photocurrent depends mainly on meteorological
conditions such as irradiation and temperature. The mathematical development conducted
in the majority of literature confirms that the irradiance has more effect on the photocurrent
and consequently on the generated current.

2.2. Proposed Artificial Neuro Networks Predictive Modeling

Forecasting allows for predicting the required future steps for different application
areas, such as commercial and industrial fields. This is mandatory for to make the necessary
decisions to improve the future situation and avoid the worst outcomes that could affect
the desired situation. Forecasting is divided into two types: qualitative and quantitative
forecasting. Time series forecasting is the most prominent of these methods. Indeed, time
series are predicted by mathematical forecasting through time-specific historical data. The
historical data are analyzed and strategic decision-making is performed for the future. For
this type of prediction, the analysis must be thorough and evidence-based to ensure that
the future outcome is achievable.
Nowadays, most technologies use artificial intelligence (AI) because of its efficiency.
Among AI techniques, those based on neural networks (NN) are being employed to address
many problems, including prediction problems. The working principle of NNs is based on
an interconnected processing element that depends on biological neurons equivalent to
pieces that carry information and transmit it to other cells in a series of networks [39].
According to Figure 3, the artificial neurons consist of three levels: the first level is the
input, which consists of a number of nodes where each node represents one of the inputs,
and the second level is the hidden level, and its number varies from one network to another
according to the level of input and output. The last level is the output level, which is the
result or the goal to be reached. All the previous levels are connected to each other through
the nodes and contain a group of nodes that receive inputs and outputs called the level,
and each node carries weights that enhance the strength of the neural connection.
 the input, which consists of a number of nodes where each node represents one of the
inputs, and the second level is the hidden level, and its number varies from one network
to another according to the level of input and output. The last level is the output level,
which is the result or the goal to be reached. All the previous levels are connected to
each other through the nodes and contain a group of nodes that receive inputs and out-
Electronics 2023, 12, 592 6 of 26
puts called the level, and each node carries weights that enhance the strength of the neu-
ral connection.

Hidden layers

Inputs layers X
Outputs layers

Tt
Tt+1

G t
Gt+1

Weights (W 1 , 2 , j )

Figure 3. Neural
Figure network
3. Neural architecture.
network architecture.

A neuron
A neuron consists
consists ofofananintegrator
integratorthat
that performs
performs the weighted
weighted sum sumofofitsitsinputs
inputs[40].
The result, P,
[40]. The result, P, of this sum is then transformed by a transfer function, Y, which pro-the
of this sum is then transformed by a transfer function, Y, which produces
output,
duces S, of theS,neuron,
the output, according
of the neuron, to the Equations
according (4) to (8).
to the Equations (4) to (8).
n
P = P
n
=∑W 1W
, j XXj −− ββ (4) (4)
1,j j
j=1
j =1

T − β
T
P P= =
WW X X−β (5) (5)

X X==[ X[ X1 1, ,XX22 ,, .....,

. . . . .,XXk ]];; W
k W == [[WW1,1 , W1,2 . . . . ., W1,n ]
1,1 , W 1, 2 ....., W 1, n ] (6) (6)
 
W1,1 W1,2 . . . . . W1,n
W = WW W
2,1 W 2,2
. . . . . WW2,n
.....  (7)
 W 1,1
W
1,2
. . . . .
1, n
W 
W =  W 2 ,1 k,1 W 2 ,2k,2 ..... W 2 , nk,n  (7)
To obtain the output of the neuron,  the activation function  Y is employed:
W k ,1 W k ,2 ..... W k , n 
S = Y ( P ) = Y (W T X − β ) (8)
To obtain the output of the neuron, the activation function Y is employed:
where S = Y ( P) = Y (W X − β ) T
(8)
X: input layers.
where
W:weights.
X: input layers.of entries, for our example, n=2.
n: numbers
W:weights.
k: number of neurons in the same layer.
n: numbers of entries,
Y: activation for our example, n=2.
equation.
k: number
β: bias. of neurons in the same layer.
Y: activation equation.
Various possible forms of the transfer function can be found. There are four popular
β : bias.
activation functions: symmetric threshold, threshold, sigmoid, and linear.
For the present case, the sigmoid function is used to forecast the outputs (T, G),
according to the following input/output relationship (9):

1
S= (9)
1 + e−(W ∗ X − β)
The NN-based prediction technique of the temperature and the irradiation (T and G)
follows these steps:
• Step1: Data assembly, pre-processing, data conversion, and normalization. The data
set used to predict the temperature and solar radiation reflected on the PV under study
was obtained from the Department of Systems Engineering and Automation at the
 S = *X −β )
(9)
1 + e − (W
The NN-based prediction technique of the temperature and the irradiation (T and
G) follows these steps:
Electronics 2023, 12, 592
• Step1: Data assembly, pre-processing, data conversion, and normalization.
7 of 26
The data set used to predict the temperature and solar radiation reflected
on the PV under study was obtained from the Department of Systems En-
gineering and Automation at the Vitoria School of Engineering of the Uni-
Vitoria School of Engineering of the University of the Basque Country. The data was
versity of the Basque Country. The data was collected using the irradiance
collected using the irradiance and temperature sensor Si-V-010-T [41].
and temperature sensor Si-V-010-T [41].
• Step2: Statistical analysis.
• Step2: Statistical analysis.
• Step3: Neural Network objects design.
• Step3: Neural Network objects design.
The Figure
The Figure 44 shows
shows the
the structure
structure of
of the
the adopted
adopted neural
neural network.
network.

Hidden

W b

20

Output

W b

Output
1

Figure
Figure 4.The
4. Theneural
neuralnetwork
networkarchitecture.
architecture.

• • Network
Step4: Step4: Network
training; training;
the algorithmthe algorithm
of Levenbergof Levenberg
Marquardt Marquardt
has been used hasfor
been
the
training used
of thefor the training
network. of the has
This choice network. This choice
been justified by the hasfact
been
thatjustified by the
this algorithm
typicallyfact that this
requires algorithm
more memory typically
but lessrequires
time. The moretraining
memory but less time.stops
automatically The
when the generalization
training stops improving,
automatically stops when as indicated
the by an increase
generalization in the mean
stops improving, as
square error of the by
indicated validation samples.
an increase in theThe
meanMean Squared
square errorError is validation
of the the averagesamples.
squared
difference
Thebetween outputs and
Mean Squared targets.
Error is theLower
averagevalues are better,
squared as zero
difference means outputs
between no error.
This algorithm is also improving the regression, R, and it is the value
and targets. Lower values are better, as zero means no error. This algorithmmeasuring the
correlation between outputs and targets. A unit, R, value indicates a close
is also improving the regression, R, and it is the value measuring the corre- relationship,
while 0 denotes a random
lation between relationship.
outputs and targets. A unit, R, value indicates a close rela-
• Step5: Simulation of network
tionship, while response
0 denotes to new
a random entries.
relationship.
• • Approval
Step6: and testing.
Step5: Simulation of network response to new entries.
The sample data process is divided into three phases: A reasonable result can be
achieved by adjusting the ANN weights during the training phase. The second phase
involves determining the minimum point of error. In the third phase, the accuracy of the
ANN is evaluated.
After presenting the proposed forecasting technique, the following section focuses
on the suggested MPPT-based control of the DC-DC chopper integrated to extract the
maximum available power independently of the meteorological conditions.
The main stages of the temperature and solar radiation prediction algorithm are
displayed in a flowchart in Figure 5.
 Best performance?

Electronics 2023, 12, 592 8 of 26

Registration and test on the best
model

Select data(define
input and output)

Read training and testining data

Choose a training algorithm

Training the network to fit the inputs

and targets

Best performance ?

Registration and test on the best model

Figure5.5.Flowchart
Figure Flowchartofofthe
thetemperature
temperatureand
andsolar
solarirradiation
irradiationprediction
predictionalgorithm.
algorithm.

Hyperparameters have a significant impact on the neural network accuracy. For the
purpose of determining the optimal set of hyper-parameters, Bayesian surrogate gaussi
an processes, gradient boost regression trees, and random forest were used [42]. The
authors in [43] presented a meta-heuristic algorithm for optimizing hyperparameters. A
cross-validation approach was used in this study to train and validate neural networks
with multiple architectures. In this method, weight optimization was performed by finding
the weights that minimized the mean squared errors (MSE) between the obtained outputs
and the desired ones, using a set of training models, and comparing the results at each
epoch against a different set of training models. In the second set, which was generally
referred to as the validation set, training was stopped once the MSE in the validation
set increases. As part of a batch training process, networks were trained using the back-
propagation algorithm with different iterations or epochs. The number of epochs and the
number of neurons in the hidden layer had a significant impact on the performance of
the neural network. Using the hidden layer neural network of two neurons and training
it for 1000 epochs, the smallest MSE validation could be obtained. In this study, 60538
temperature and solar radiation values were collected, of which 70% were recorded for
training and 30% were used for validation and testing. With 20% of the hidden layer’s size,
this was achieved.
In this network, two layers of feed-forward neurons with sigmoid hidden neurons
and linear output neurons were used to perform regression tasks. The neural network was
trained using the Levenberg-Marquardt algorithm, in which training was automatically
stopped when there was no further improvement in generalization. An analysis of the
mean square error (MSE) and regression (R) indicated the completeness of this algorithm.
According to the results of this study, MSE = 0.003 and R = 0.998 were found, indicating a
strong correlation between outputs and objectives.
Concerning the choice of input features, various research works have dealt with this
subject [44]. In the present work, each additional neuron allows for the consideration
of specific profiles of the input neurons. A larger number, therefore, allows for it to be
possible to better adhere to the data presented but reduces the generalization capacity
of the network. At this time, there is no general rule but rules of thumb. A subsequent
 rithm. According to the results of this study, MSE = 0.003 and R = 0.998 were found, in-
dicating a strong correlation between outputs and objectives.
Concerning the choice of input features, various research works have dealt with
this subject [44]. In the present work, each additional neuron allows for the considera-
Electronics 2023, 12, 592 tion of specific profiles of the input neurons. A larger number, therefore, allows for9itofto26
be possible to better adhere to the data presented but reduces the generalization capacity
of the network. At this time, there is no general rule but rules of thumb. A subsequent
research path would consist of estimating a network comprising of many neurons and
research path would consist of estimating a network comprising of many neurons and
then in simplifying it by analyzing multi-collinearities, by learning rule eliminating use-
then in simplifying it by analyzing multi-collinearities, by learning rule eliminating useless
less neurons, or by defining an architecture considering the structure of the variables
neurons, or by defining an architecture considering the structure of the variables identified
identified beforehand by a principal component analysis.
beforehand by a principal component analysis.
2.3.
2.3.JAYA-SMC
JAYA-SMC Hybrid
Hybrid MPPT
MPPT Control
Control of
of the
the SEPIC
SEPIC Chopper
Chopper
2.3.1.
2.3.1.Integrated
IntegratedSEPIC
SEPICChopper
Chopper
The
The single-ended
single-endedprimary-inductor
primary-inductorconverter
converter(SEPIC)
(SEPIC)isisaamodification
modificationof ofthe
the Basic
Basic
Boost
Boost and
and Cuck Converter.
Converter. It It is
is more
moreperformant
performantthanthanother
otherDC-DC
DC-DCconverters
converters inin terms
terms of
of purity
purity andand efficiency
efficiency of the
of the input input current
current sincesince it shows
it shows verybypass
very little little bypass or ringing,
or ringing, as well
as well
as a reduced switching
as a reduced loss. The
switching output
loss. noise and
The output power
noise and phase
powerthatphasecanthat
be operated
can be oper-at a
muchathigher
ated a much frequency than that ofthan
higher frequency otherthat
inverters willinverters
of other also be reduced.
will alsoThebe output
reduced.voltage
The
achieved
output by theachieved
voltage SEPIC converter
by the SEPICis non-inverting.
converter isFigure 6 shows the
non-inverting. electrical
Figure diagram
6 shows the
of a SEPIC
electrical converter.
diagram of a The
SEPICdifferent values
converter. of different
The the electronic
valuescomponents are mentioned
of the electronic compo-
in Appendix
nents B.
are mentioned in Appendix B.

Figure 6. Electrical diagram of SEPIC converter.

The SEPIC chopper operation principle is analyzed in two stages according to the
conduction state of the switch K. Based on the notations of the Figure 6, when the switch
K is closed, the behavior of the SEPIC is described by Equation (10). When it is open,
Equation (11) describe the behavior of the SEPIC.
dil1 1


 dt = L1 · Ve
 dil2 1
= L2 · Vc1


dt
dVc1 (10)

 dt = C11 · i L1

dVc2
= − C12 · Vc2


dt

dil1 1 1
  

 dt = L1 · Ve − L1 · VC1 + VC2
dil2 1
dt = − L2 · VC2



dVC1 1 (11)
dt = − C1 · i L2



 dVC2 1 1
dt = C2 · (il1 + il2 ) − C2 · VC2


The state model of the converter is inferred from Equations (10) and (11). This model
is presented in Equation (12).
dil1 1 1
L1 · Ve − L1 · (1 − d ) · [Vc1 + Vc2 ]


 dt =
dil2
= Ld2 · Vc1 − (1L−2d) · Vc2



dt
dVc1 (12)


 dt = (1C−1d) · il1 − Cd1 · i L2
dVc2
= (1C−2d) · (il1 + il2 ) − C12 · Vc2


dt

The parameter d refers to the duty cycle of the SEPIC chopper. Such a parameter is
mainly determined via the maximum power point tracking technique.
Electronics 2023, 12, 592 10 of 26

2.3.2. JAYA-SMC Hybrid MPPT Control

One of the most important problems that the PV generation faces is the problem of
maximum power tracking ability, which is mandatory for the DC-DC chopper. In fact,
the solution exists to control such a converter to take maximum advantage of the power
produced by the PV panels. In order to overcome the drawbacks of the conventional MPPT
methods, such as perturb and observe (P&O), incremental conductance (IC), etc. ..., a recent
MPPT based on the hybridization of the JAYA optimization algorithm and the sliding mode
control (SMC) is proposed. This MPPT technique is employed to maximize the power
generated under the constraints of equality that includes the relationship between current
and voltage. This problem aims to raise the PV power by setting the optimum value of
I and V, where the equations expressing the optimization problem and the constraint of
equality R (I, V) = 0 are established in Equations (13) and (14):
(
max[ P( I, V )]
I,V (13)
R( I, V ) = 0

With (
P( I, V ) = I · V
q(V + IRs )
V + IRs (14)
R( I, V ) = I ph − I0 (e aKTNs − 1) − Rsh −I
The objective is to solve the optimization problem based on equation in (13).
1. Jaya Method
Heuristic algorithms are becoming a good solution to model-free optimization prob-
lems [45–47]. In this article, an advanced swarm-based JAYA algorithm is used to find
the maximum power point for PV arrays. This algorithm does not require any specific
parameters to be configured. Therefore, it is easy to implement without having to modify
the initialization parameters.
In the present work, f (x) is considered as the objective function to be maximized as
the power P(X) = P (V, I). The JAYA algorithm searches Vmax and Imax allowing it to reach
Pmax . It is easy to implement, in each iteration of i, as it is assumed that there is a number of
design variables, m, (j = 1, 2, m), and n, the number of candidate solutions that determine
the community size, K, (k = 1, 2, n). The best candidate obtains the best value of P(X) in the
candidate solution set and the worst candidate obtains the worst value of P(X) in the set of
candidate solutions. If X(i,j,k) is the value of the variable, jth , of the filter Kth, in iteration ith
this value is adjusted in Equation (15).

Xn (i, j) = X (i, j, k) + r1,i,j ( Xbest ( j) − | X (i, j)|) − r2,i,j ( Xworst ( j) − | X (i, j)|) (15)

Herein, X(i,j) is jth from the assumption of the solution ith , |X(i,j)| is the absolute value
of X(i,j), Xbest(j) is the best solution, Xworst(j) is the worst solution, Xn(i,j) is the update of
the variable X(i,j), andr1,2 are two random numbers belonging to the interval [0,1].
The main steps of Algorithm 1 are illustrated in the following description.
Electronics 2023, 12, 592 11 of 26

Algorithm 1: JAYA Algorithm

Step 1: Set the population and the maximum number of iterations NPop and
Nmax.
Step 2: Determine the Xbest andXworst solutions.
Step 3: While gen <= ≤ Nmax
For I = 1 to Npop carry out:
Xn (i, j) = X (i, j, k) + r1,i,j ( Xbest ( j) − | X (i, j)|) − r2,i,j ( Xworst ( j) − | X (i, j)|)
Obtain the update community and evaluate the new value, if the new value is
more suitable than the previous one, it will replace the old one.
End for; End while.
Step 4: Show existing solutions X(i) and f(X(i)).

Electronics 2023, 12, 592 It is important to recall that the JAYA algorithm is used to optimize the parameters
12 of of
28
the sliding mode control.
2. Sliding Mode Control Technique
To determine the
To determine the duty
duty cycle
cycle ofof the
the SEPIC
SEPIC chopper,
chopper, the the sliding
sliding mode
mode control
control (SMC)
(SMC)
has been adopted.
adopted. The The SMC
SMC isisaakind
kindofofvariable
variablestructure
structurecontrol
control algorithm.
algorithm. The
The fun-
funda-
damental difference
mental difference between
between sliding
sliding modemode control
control andand conventional
conventional control
control strategy
strategy is
is the
the discontinuity of its control [48]; that is, the controller output changes over
discontinuity of its control [48]; that is, the controller output changes over time and exhibits time and
exhibits
switchingswitching characteristics.
characteristics. Since the
Since the sliding sliding
mode mode is independent
is independent of system and
of system parameters pa-
rameters and the
disturbances, disturbances,
sliding mode the sliding
system hasmode
strongsystem has strong
robustness. robustness.
The basic principleThe basic
of sliding
principle of sliding
mode is shown mode7,iswhere
in Figure shownthe in Figure
following7, where the processed:
steps are following steps are processed:
• First,• design
First,adesign
slidingasurface
slidingin surface in state space.
state space.
• •
Have a selection of a control law to forcetothe
Have a selection of a control law force thetrajectory
state state trajectory of the system
of the system to moveto
move towards a predetermined
towards a predetermined surface in finite time. surface in finite time.
• • Maintain
Maintain around thisaround thiswith
surface surface with appropriate
appropriate switching switching
logic. logic.

X2

X(t0)

Sliding surface Reachinig phase

X1

X(t1)
Sliding surface

S=0

Figure 7. SMC basic principle.

Figure 7.SMC basic principle.
In order to regulate the converter input current, the following sliding surface is chosen:
In order to regulate the converter input current, the following sliding surface is cho-
sen: S = ire f − i L1 (16)
S = i f − iL1 (16)
where iref is the optimum current ofr ethe MPPT control and iL1 the input current of the
converter
Where iref SEPIC. By defining
is the optimum the above
current of thesurface,
MPPT the control
control andlaw
iL1 should
the input be current
applied of
to the
the
converter to force the system to move on the sliding mode surface in a finite time,
converter SEPIC. By defining the above surface, the control law should be applied to the according
to the following structure for control input:
converter to force the system to move on the sliding mode surface in a finite time, ac-
cording to the following structureUfor (t)control
= U (input:
t) + U (t) (17)
eq n
U (t ) = U eq (t ) + U n (t ) (17)
Where Ueq defines the system’s behavior on the sliding surface and is known as the
equivalent control-input, Un is the nonlinear switching input that moves the state to the
sliding surface and maintains the state on such surface in the presence of the uncertain-
Electronics 2023, 12, 592 12 of 26

where Ueq defines the system’s behavior on the sliding surface and is known as the equiva-
lent control-input, Un is the nonlinear switching input that moves the state to the sliding
surface and maintains the state on such surface in the presence of the uncertainty. Ueq is ob-
.
tained from the invariance condition and is presented as below: (S = 0, S = 0) ⇔ U = Ueq
. . .
S = ire f − i L1 = 0
.
1 1
(18)
ire f − L1 · Ve + L1 · (1 − Ueq ) · [Vc1 + Vc2 ] = 0

Vc1 + Vc2 − Ve
Ueq = (19)
[Vc1 + Vc2 ]
.
Here, Un is chosen so that the Lyapunov stability criteria (V < 0) is met.
Where
1
V = S2 (20)
2
. .
V=  S·S < 0  (21)
S · − L1 · Ve + L1
1
· ( 1 − U ) · [Vc1 + Vc2 ] < 0
1
 
1 1
S · − · Ve + · (1 − Ueq − Un ) · [Vc1 + Vc2 ] < 0 (22)
L1 L1
[Vc +Vc ]
− 1L1 2 · Un · S < 0 (23)
⇒ S · Un > 0
The nonlinear component is given by:

Un = K. sign(S), (24)

+1 si S > 0
where K > 0; sign(S) = .
−1 si S < 0
The gain K is chosen to be positive. The choice of this gain is very influential since
if it is small, the controller loses the robustness properties and if it is large, important
oscillations will be derived at the level of the control unit. These oscillations can excite
the neglected dynamics (Chattering phenomenon), or even damage the control unit [49].
Chattering can be reduced by replacing the “sign” function with a hyperbolic tangent
function (tanh) [50]. The Equation (25) represents the new form of the command law:

1
U= (1 + tanh(s)) (25)
2

3. Simulation Results and Discussion

The studied system depicted in Figure 1 integrates the PV generator, the DC-DC
chopper, and the DC load. To simulate a more intricate consumer, the selected load was a
DC motor. The Appendix A, Appendix B, and Appendix C summarize, respectively, the
characteristics of the peimar SG340P commercial PV panel, the DC-DC converter, and the
DC motor used to carry out the simulation.
The first simulations have been focused on the SG340P PV panel behavior when faced
with different meteorological conditions of temperature and irradiance. In this context,
Figures 8 and 9 represent the I-V and P-V characteristics of the adopted PV panel, under
different temperatures and irradiations.
 Electronics 2023, 12, 592 15 of 29
Electronics 2023, 12, 592 15 of 29
Electronics 2023, 12, 592 13 of 26

Figure 8. PV panel I-V characteristic.

Figure 8. PV
Figure 8. panel I-V I-V
PV panel characteristic.
characteristic.

Figure
Figure 9. PV
9. PV panel
panel P-V
P-V characteristic.
characteristic.
Figure 9. PV panel P-V characteristic.
ToTo test
test the
the effectiveness
effectiveness of of
thethe JAYAalgorithm,
JAYA algorithm, this
this method
method has
has been
been applied
applied with
with
To
variabletest the effectiveness
variabletemperature
temperatureand of
andsolarthe JAYA
solar radiation.algorithm,
radiation. Figure
Figure10 this
10showsmethod
showsthe has been
variable
the applied
testtest
variable with
conditions used.
conditions
variable
used. temperature and solar radiation. Figure 10 shows the variable test conditions
used.
 Electronics 2023,
Electronics 12, 592 1514ofof2826
Electronics 2023, 12,12,
2023, 592592 15 of 28

(a)
(a) (b)
(b)
Figure10.
Figure 10.Solar
Solarirradiance
irradianceand
andtemperature
temperature
Figure 10. variation :(a)Solar
Solar irradiance
variation :(a) Solar
and irradiance variation…(b)
temperature
irradiance Temperature
variation: Temperature
variation…(b) (a) variation….
Solar irradiance variation, (b) Temperature
variation….
variation .
Figure1111indicates
Figure indicatesthat,
that,ininmost
mostcases,
cases,the thepower
powercurve
curveisissmaller
smallerwhen
whencompared
compared
totothe
themaximum
maximum
Figure value.The
11 indicates
value. The
that,power
in most
power curve is,is,the
cases,
curve ininmost
mostcases,
power cases, smaller
curvesmaller
is smallerthan
than themaximum
when
the maximumto
compared
value.
the Despite
maximum abrupt
value. changes,
The power the MPP
curve detection
is, in most algorithm
cases, performs
smaller than
value. Despite abrupt changes, the MPP detection algorithm performs satisfactorily. Due satisfactorily.
the maximum Due
value.
to the complexity
Despite abrupt of calculations,
changes, the MPP this is the
detection case. Indeed,
algorithm there
performs is an inherent
to the complexity of calculations, this is the case. Indeed, there is an inherent limitationthe
satisfactorily. limitation
Due to
totocomplexity
othermethods
other methods ininthis
thisregard.
regard.
of calculations, this is the case. Indeed, there is an inherent limitation to other
methods in this regard.

Figure11.
Figure 11.PV
PVtesting
testingofofthe
theoptimization
optimizationalgorithm..
algorithm..
Figure 11. PV testing of the optimization algorithm.

AtAtAt this
thisthis stage,
stage,
stage, the
thethe main
main
main objective
objective
objective isisto
istoto assess
assess
assess the
thethe performance
performance
performance and
and
and accuracy
accuracy
accuracy ofofofthe
thethe
forecasting
forecasting
forecasting system.
system.
system. For the
Forthe
For training,
thetraining,
training,thethe errors
theerrors were repeated
errors were repeated 6
repeated 66times times
timesafterafter
after200 200 epochs
200epochs
epochs and
and
and the the
thetest test
testwas was
was stopped
stopped
stopped atat atepoch
epoch
epoch 212 212
212withwith
with agradient
gradient
a agradient of0.09.
0.09.
ofof0.09. The The
The error
error
error isrepeated
repeated
isisrepeated fromfrom
from the
the
the 200
200th
200
th era, which showed the growing importance of data. Therefore, age 200 is cho-
th era,
era, which
which showed
showed the the growing
growing importance
importance of of data.
data. Therefore, age 200 is chosen cho-
sen
sen asthe
asasthe the
base base
base and and
andthe the weights
weights
the weights are
areare chosen
chosen
chosen as as asthe
the thefinal
finalfinal weights.
weights.
weights. InInaddition,
addition,
In addition, thevalida-
thethe valida-is
validation
tion
tion isis212
6 to 6 6toto 212epochs,
epochs,
212 epochs,
sincethesincethe errors
errorserrors
sincethe arerepeated
are repeated
are repeated 6 6times
6 times timesbefore
before before theprocess
the process
the process stopsde-
stops stops de-
definitively
finitively
(see Figure (see Figure
12).
finitively (see Figure 12). 12).
Electronics 2023, 12, 592 16 of 28

Electronics 2023, 12, 592 16 of 28

Electronics 2023, 12, 592 15 of 26

Figure 12. ANN training.

In Figure 13, the regression and its values measure the correlation between outputs
and targets are presented. The R value is almost equal to 1, which means a close rela-
Figure 12. ANN training.
tionship.
Figure 12. ANN training.
Figure 14 illustrates that the MSE approaches 0 for the training and validation loss-
In Figure 13, the regression and its values measure the correlation between outputs and
In Figure
es at epoch 13, graph
10. The the regression and its values
below illustrates measure theofcorrelation
the performance between
the proposed outputs
method.
targets are presented. The R value is almost equal to 1, which means a close relationship.
and targets are presented. The R value is almost equal to 1, which means a close rela-
tionship.
Figure 14 illustrates that the MSE approaches 0 for the training and validation loss-
es at epoch 10. The graph below illustrates the performance of the proposed method.

FigureFigure ANN regression.

13. regression.
13. ANN

Figure 14 illustrates that the MSE approaches 0 for the training and validation losses
at epoch 10. The graph below illustrates the performance of the proposed method.
Figure 13. ANN regression.
Electronics 2023, 12, 592 17 of 28
Electronics 2023,
Electronics 12,12,
2023, 592592 17 of
16 28
of 26

Figure 14. ANN performance.

Figure 14.14.
Figure ANN performance.
ANN performance.
Figure 15 represents the ANN forecasting model on Simulink.
Figure
Figure 15 15 represents
represents the ANNtheforecasting
ANN forecasting
model model on Simulink.
on Simulink.

Figure
Figure 15. Simulink
15. Simulink modelmodel of ANN
of ANN forecasting.
forecasting.
Figure 15. Simulink model of ANN forecasting.
According
According to the
to the ANN ANN forecasting
forecasting model
model of of Figure
Figure 14,14,the
theresults
resultsfound
foundafter
after the
the test
of According
neurons aretoshown
the ANNin Tables 1 and
forecasting 2:
test of neurons are shown in Tables 1 and 2:model of Figure 14, the results found after the
test-The
of neurons
inputs are shown in
are theforecast
last threeTables 1 and 2:temperatures of the PV panels (Tk, Tk-1, and
previous
Table 1. Temperature testing.
Tk-2). -The inputs are the last three previous temperatures of the PV panels (Tk, Tk-1, and
Tk-2-The
). real temperatureReal
Inputs is the temperature found
Temperature in the real-time.
Forecasted Temperature Error
-The
-The real temperature
forecasted temperature is theistemperature
the forecastedfound in the real-time.
temperature found by the neural
(54.43;54.50;54.54) 54.59 54.58 0.01net-
work-based-The forecasted temperature
technique (Tk+1).
(26.15;26.15;26.06) 25.96
is the forecasted temperature
26.00
found by the neural
0.04
net-
work-based technique
(59.86;59.76;59.91) (T k+1 ). 60.08 60.09 0.01
Table 1. Temperature forecast testing.
Table 1. Temperature forecast testing.
Inputs Real temperature Forecasted Temperature Error
Inputs Real temperature Forecasted Temperature Error
(54.4354.50;54.54) 54.59 54.58 0.01
(54.4354.50;54.54)
(26.15;26.15;26.06) 54.59
25.96 54.58
26.00 0.01
0.04
(26.15;26.15;26.06) 25.96 26.00 0.04
 Electronics 2023, 12, 592 17 of 26

Electronics 2023, 12, 592 Table 2. Irradiation forecast testing. 18 of 28

Input Real Irradiation Forecasted Irradiation Error

(959.28;960.38;961.85)
(59.86;59.76;59.91) 962.58
60.08 962.71
60.09 0.13
0.01
(548.03;547.66;546.75) 545.65 546.1 0.45
(877.25;877.80;877.80) 876.70 877.38 0.68
Table 2. Irradiation forecast testing.

Input
-The inputs are the Real Irradiation
last three previous Forecasted Irradiation
temperatures Error (T , T ,
of the PV panels k k−1
(959.28;960.38;961.85)
and Tk−2 ). 962.58 962.71 0.13
(548.03;547.66;546.75)
-The real temperature is545.65
the temperature found546.1 in the real-time. 0.45
(877.25;877.80;877.80)
-The forecasted temperature876.70 877.38
is the forecasted temperature 0.68 network-
found by the neural
based technique (T
The proposed neural ).
k+1 network-based forecasting technique is tested on the real data
The proposed
and comparisons neuralthe
between network-based
real and the forecasting technique is tested
predicted temperature, onas
as well thethe
real data and
irradia-
comparisons
tion are conductedbetween
withthethereal
aimand the predicted
of reducing temperature,
the error betweenasthewell as theand
target irradiation
the fore-are
conducted
casted values.with the aim of reducing the error between the target and the forecasted values.
The used MPPT −2 seconds; depending on its output the
The used MPPT hashas a sampling
a sampling time
time ofof
1010
−2 seconds; depending on its output the
sliding mode controller has been used to find the
sliding mode controller has been used to find the duty cycle. duty cycle.
According
According to to
thethe
samesame principle
principle of using
of using thethe
lastlast previous
previous three
three values
values of tempera-
of tempera-
ture and irradiance to deduce their forecasted values, Figure 16 represents
ture and irradiance to deduce their forecasted values, Figure 16 represents the results the results
of of
temperature and irradiation forecasting.
temperature and irradiation forecasting.

(a) (b)
Figure 16. Forecasting results :(a) Temperature forecast…(b) Irradiation forecast….
Figure 16. Forecasting results: (a) Temperature forecast, (b) Irradiation forecast.

From
FromFigure
Figure 16,16,ititcan
canbe
be observed
observed thatthat there
thereisisaagood
goodagreement
agreement between
between thethe ex-
expected
pected
valuesvalues
and theandreal
theones.real ones.
As canAsbecan be seen,
seen, the SEPIC
the SEPIC duty cycle
duty cycle is adjusted
is adjusted in ac-
in accordance
cordance
with thewith the changes
changes in the MPPT.in the MPPT.
Thus, Thus, it demonstrates
it demonstrates its ability toitsadapt
ability to adaptintothe
to changes
changes
systeminasthetheysystem
occur.asAs they occur.
shown As shown
in Figure in Figure
17, the difference17, between
the difference betweencurrent
the reference the
reference
(Impp ) and the input current of the SEPIC is approximately 10%. It is pertinent to note is
current (Impp ) and the input current of the SEPIC is approximately 10%. It that
pertinent to is
this error note
verythat this which
small, error is very small,
indicates which indicates
the accuracy of the SMC.the accuracy
According of to
the
theSMC.
graph,
According
there aretosomethe graph,
peaks causedthere are
by some
changes peaks caused
in the by changes
reference current. in These
the reference
changescur- occur
rent. These
within 25 changes
millisecondsoccurwithout
within causing
25 milliseconds without
any problems forcausing any problems for the
the system.
system.
Figures 17 and 18 show the controller signal and current error. This signal repre-
sents the duty cycle of the SEPIC converter which is limited between 0 and 1.
ronics 2023, 12, 592 19 of 28
Electronics 2023, 12, 592 18 of 26

0.9

0.8

0.7

0.6

0.5

0.4

0.3

0.2
0 2 4 6 8 10 12 14 16 18 20
Time(s)

Figure 17. Controller signal.

Figures 17 and 18 show the controller signal and current error. This signal represents
the duty cycle of the SEPIC converter which is limited between 0 and 1.
The previous model is applied to predict the temperature and the irradiation, while
the found responses are depicted in the following figures.
In accordance with Figures 19 and 20, showing the temporal evolution of the DC
motor current and voltage, respectively, Figure 20 represents the power of such a motor.
The power is less than the PV generator power, since at the SEPIC output, the current
is oscillating: the load (plus its filtering capacitor) is only supplied for a fraction of the
cycle. This requires the use of large filter elements, which are, moreover, highly stressed.
Electronics 2023, 12, 592 20 of 28
Figure 21 represents the speed of the DC motor. As the voltage and current change, the
speed changes as well (See Figure 22).

Figure 17. Controller signal.

Figure 18. Current error.

Electronics 2023, 12, 592 21 of 28
Electronics 2023, 12, 592 21 of 28
Electronics 2023, 12, 592 19 of 26

Electronics 2023, 12, 592 21 of 28

Figure 19. Temporal evolution of the motor current.

Figure 19. 19.
Figure Temporal evolution
Temporal of the
evolution motor
of the current.
motor current.
Figure 19. Temporal evolution of the motor current.

Figure 20. Temporal evolution of the motor voltage.

Figure 20. 20.
Figure Temporal evolution
Temporal of the
evolution motor
of the voltage.
motor voltage.
Figure 20. Temporal evolution of the motor voltage.

Figure 21.Temporal
Figure evolution
21. Temporal of the motor
evolution of thepower.
motor power.

Figure 21.Temporal evolution of the motor power.

Figure 21.Temporal evolution of the motor power.
Electronics 2023, 12, 592 22 of 28
Electronics 2023, 12, 592 22 of 28
Electronics 2023, 12, 592 20 of 26

Figure 22.Temporal evolution of the motor speed.

Figure 22.Temporal
Figure evolution
22. Temporal of the
evolution motor
of the speed.
motor speed.
The previous figures show the various changes in the DC motor behavior according
to theThe
predicted
Theprevioustemperature
previous figures
figures showandthe
show irradiation
the various
various profiles.
changes
changes These
in in
thethechanges
DCDC are
motor
motor due to according
behavior
behavior the MPP
according
location
to the changing.
predicted When the
temperature predictor
and reaches
irradiation the predicted
profiles. These value, the
changes
to the predicted temperature and irradiation profiles. These changes are due to the MPP algorithm
are due to finds
the MPP
the MPP
location
location point, so
changing. the
changing.When curves
Whenthe join the
thepredictor steady
predictorreachesstate.
reaches the
the predicted
predicted value,
value, the
thealgorithm
algorithmfinds
findsthe
In Figures
theMPP
MPP point,so23–25
point, sothe at
thecurvescertain
curves join moments,
jointhe
thesteady the
steadystate.current, voltage, and power exceed the
state.
maximumIn In value,23–25
Figures
Figures which
23–25 atindicates
at certain
certain anmoments,
increase
moments, theor
the decrease.
current,
current, This isand
voltage,
voltage, due
and topower
powerthe exceed
change
exceed inthe
the
radiation and temperature at these moments, which indicates that
maximum value, which indicates an increase or decrease. This is due to the change in in
maximum value, which indicates an increase or decrease. Thisthe
is power
due to has
the not
changeyet
radiation
reached
radiation theandand temperature
maximum
temperaturevalue. at
atThe these
these moments,
response
moments, which
stabilizes
which at aindicates
maximum
indicates that
that ofthe
the 2.5 power
powerms, hashas
notnot
which ayet
isyet
reached the maximum value. The response stabilizes at a maximum of 2.5 ms, which is ais a
reached
reasonable the maximum
value. value. The response stabilizes at a maximum of 2.5 ms, which
reasonable
reasonable value.
value.

Figure
Figure 23. Temporal
23.Temporal evolution
evolution of current
of PV PV current
overover time.
time.
Figure 23.Temporal evolution of PV current over time.
Electronics 2023, 12, 592 23 of 28
Electronics 2023, 12, 592 21 of 26

Electronics 2023, 12, 592 23 of 28

Figure 24.Temporal
Figure evolution
24. Temporal of PV
evolution of voltage overover
PV voltage time.
time.
The analysis of the dynamic behavior of the current, voltage, and generated power-
highlights the success
Figure 24.Temporal to follow
evolution the MPP
of PV voltage point
over time.and, therefore, confirm the effectiveness of
the proposed JAYA-SMC to find the maximum available power.

Figure 25. Temporal evolution of the PV generator power.

The analysis of the dynamic behavior of the current, voltage, and generated pow-
Figure
Figure 25.25. Temporal
Temporal evolution
evolution of of
thethe
PVPV generator
generator power.
power.
erhighlights the success to follow the MPP point and, therefore, confirm the effectiveness
of the proposed
Following JAYA-SMC to find the discussion
maximum of available power.
The analysistheof presentation
the dynamicand behavior of the the main
current, achievedand
voltage, results, Table 3pow-
generated briefly
Following
shows a the presentation
comparative analysis and
thatdiscussion
has been of the main
developed. achieved
This results, Table
comparative study 3has
erhighlights the success to follow the MPP point and, therefore, confirm the effectiveness
briefly shows on aworks
comparative analysis that the
has subject
been developed. This comparative study
offocused
the proposed that have
JAYA-SMC dealtthe
to find with
maximum of prediction
available power. using different feasible
hasapproaches.
focused on worksThe that have dealt
comparison is withon
based thethe
subject of prediction
studied system using
and different
hasresults,
included feasi-
the use
Following the presentation and discussion of the main achieved Table 3
bleofapproaches.
the The model,
prediction comparison
the is based
main on theofstudied
objective the system
study, the and hasof
degree included the use
complexity of the
briefly shows a comparative analysis that has been developed. This comparative study
of the prediction
concerned model,and
approach, the other
main performance
objective of the study, themetrics.
degree of complexity of the
has focused on works that have dealt with theand accuracy
subject of prediction This
usingcomparative study
different feasi-
concerned
highlights approach,
the and other of
competitiveness performance
the proposed and accuracy
method by metrics. Thisit comparative
benchmarking against recently
ble approaches. The comparison is based on the studied system and has included the use
study highlights the competitiveness of the proposed method by benchmarking it
ofpublished techniques.
the prediction model, the main objective of the study, the degree of complexity of the
against recently published techniques.
concerned approach, and other performance and accuracy metrics. This comparative
study highlights the competitiveness of the proposed method by benchmarking it
Table 3. Comparative study.
against recently published techniques.
Ref. Studied Model Main Objective Degree of MSE R2
Table 3. Comparative study.

Ref. Studied Model Main Objective Degree of MSE R2

Electronics 2023, 12, 592 22 of 26

Table 3. Comparative study.

Degree of
Ref. Studied System Model Main Objective MSE R2
Complexity
Short-term energy forecasting
[51] PVS QSVM for building integrated PV High 0.16 0.88
system
PVS Decision Tree High 0.087 0.88
Solar
Midterm solar radiation
[52] radiation-based CNN-BiLSTM High 0.17 0.94
prediction
power plants
Meteorological Estimation of daily solar
[53] DL High 0.6 0.98
ground stations radiation
[54] Meteorologicalstation ANFIS Predict solar radiation Low 1.16 0.85
Estimation of daily global
[55] SES GMDH Medium 0.05 0.98
solar radiation
Monthly solar radiation
[56] SES ANFIS-PSO High 0.09 0.99
prediction

The ANN proposed model has been compared to the available literature: QSVM,
Decision Tree [51], CNN-BiLSTM [52], DL [53], ANFIS [54], GMDH [55], and ANFIS-
PSO [56] that have been proposed for many varieties of solar radiation prediction. The MSE
and R2 criteria were used for an accurate comparison. According to Table 3, the results
clearly showed that the ANN model performs more accurately than the mentioned models
according to the values of MSE = 0.003, and R2 = 0.98. In addition, these studies are difficult
to implement due to the issues related to the identification of hyper-parameters and specific
parameters, which is not the case for the proposed approach.

4. Conclusions and Future Works

In this paper, we propose the application of artificial neurons and JAYA-SMC to
predict, control, and research the maximum power of photovoltaic panels. For predicting
the temperature and solar radiation on the PVS, artificial neural networks have been used.
It is worth noting that this method is considered to be the most performant in this field due
to its effectiveness. This is because it has hidden layers that provide it with a competitive
advantage when it comes to predictive analysis. As a consequence, it is more accurate. As
opposed to deep learning, it does not require external instructions when an error occurs
or when an incorrect prediction is made. A mean square error of 0.03% and a regression
coefficient of 0.98 were achieved with the proposed method. Based on these results, we can
see that the target and the real output have a high degree of convergence and correlation.
After that, JAYA-SMC was used to search and control the maximum power of the PVS
by searching the MPP based on JAYA’s algorithm. Additionally, the proposed method
eliminates the inconveniences associated with OCV and SCC techniques, as well as the
necessity of additional fixed keys and additional measurements. Since photovoltaic systems
are dependent on converters so they can be connected to loads, the SEPIC duty cycle was
controlled by the SMC by controlling the input current and using the Imp provided by JAYA
as a reference. Simulation results demonstrate the effectiveness of the SMC. The current
and voltage curves of the MPPT controller and PV are shown to be matched, indicating the
controller’s effectiveness.
In spite of the highly developed advanced forecasting method used in this work, the
results still need to be improved. This is since this method has not been experimentally
tested and has only been evaluated through simulation tests. There are several limitations
associated with this method, such as storage and diagnosis, especially when the system is
connected to other sources of energy. Due to advances in artificial intelligence, it can be
used to diagnose and predict energy consumption, as well as to solve the storage problem
faced by renewable energy sources.
Electronics 2023, 12, 592 23 of 26

Future work should include the integration of energy storage into the studied system
to address the issue of PV solar source fluctuations. It is one of the most promising
energy storage technologies to incorporate AEMFCs, since they combine the advantages of
PEMFCs and the economics of alkaline batteries. Such an application would be a significant
achievement that would enable the economic adoption of hydrogen energy storage systems
throughout the world.

Author Contributions: O.B.: conceptualization, simulation, data collection, and review; A.K.-M.:
simulation and data collection; M.J.: writing original draft; M.J., R.A., and H.J.: writing, review, and
editing; F.H. and M.A.: supervision. All authors have read and agreed to the published version of
the manuscript.
Funding: This study received no external funding.
Data Availability Statement: Not applicable.
Acknowledgments: The authors wish to express their gratitude to the Basque Government, through
the project EKOHEGAZ (ELKARTEK KK-2021/00092), to the Diputación Foral de Álava (DFA),
through the project CONAVANTER, and to the UPV/EHU, through the project GIU20/063, for
supporting this work.
Conflicts of Interest: The authors declare no conflict of interest.

Abbreviation List

ANN Artificial Neural Network

AI Artificial intelligence
NN Neural networks
R Regression
MSE Mean squared error
MPP Maximum power point
MPPT Maximum power point tracking
P&O Perturb and observe
Tanh Hyperbolictangent
OCV Open-circuit voltage
SCC Short-circuit current
PVS Photovoltaic system
SEPIC Single ended primary inductor converter
SMC Sliding mode control
mFFO Modified fire-fly optimizer
FE-SVR Feature engineering-support vector regression
FLC Fuzzy logic control
GA Genetic algorithm
PSO Particle swarm optimization
CPSO Chaotic particle swarm optimization
GWO Grey wolf optimization
PI Proportional integral
DC Direct current
PID Proportional integral derivative
PN Positive negative
IC Incremental conductance
SES Solar energy system
QSVM Quadratic support vector machine
CNN-BiLSTM Convolution neural network-bi-direction long short term memory
ANFIS Adaptive neuron fuzzy inference system
GMDH Group method of data handling
ANFIS-PSO Adaptive neuron fuzzy inference system-particle swarm optimization
Electronics 2023, 12, 592 24 of 26

Appendix A

Table A1. PV panel parameters.

Peak Power (Pmax) 340 W

Voltage at Pmax (Vmp) 36.7 V
Current at Pmax (Imp) 9.28 A
Open circuit voltage (Voc) 45.2 V
Short circuit current (Isc) 9.9 A

Appendix B

Table A2. Converter parameters.

Frequency
55 (KHz) L1 1.8 (mH) L2 1.4 (mH)
PWM
C1 120 (µF) C2 470 (µF)

Appendix C

Table A3. DC motor parameters.

Armature Resistance Ra 2.2 Ω

Armature inductance La 5 × 10−3 H
Back-emf constant 0.015 V/rmp
Total inertia J 0.03 kg.m2
Viscous friction coefficient 0.12 N.m.s
Coulomb friction torque Tf 0.11 N.m
Initial speed 3 rad/s

References
1. Yang, Y.; Sun, S. Tourism demand forecasting and tourists’ search behavior: Evidence from segmented Baidu search volume. Data
Sci. Manag. 2021, 4, 1–9. [CrossRef]
2. Zhidan, L.; Wei, G.; Qingfu, L.; Zhengjun, Z. A hybrid model for financial time-series forecasting based on mixed methodologies.
Expert Syst. 2021, 38, e12633. [CrossRef]
3. Gourvenec, S.; Sturt, F.; Reid, E.; Trigos, F. Global assessment of historical, current and forecast ocean energy infrastructure:
Implications for marine space planning, sustainable design and end-of-engineered-life management. Renew. Sustain. Energy Rev.
2022, 154, 111794. [CrossRef]
4. Acharya, S.; Young-Min, W.; Jaehee, L. Day-ahead forecasting for small-scale photovoltaic power based on similar day detection
with selective weather variables. Electronics 2020, 9, 1117. [CrossRef]
5. Ellahi, M.; Usman, M.R.; Arif, W.; Usman, H.F.; Khan, W.A.; Satrya, G.B.; Daniel, K.; Shabbir, N. Forecasting of Wind Speed and
Power through FFNN and CFNN Using HPSOBA and MHPSOBAACs Techniques. Electronics 2022, 11, 4193. [CrossRef]
6. Boretti, A. Integration of solar thermal and photovoltaic, wind, and battery energy storage through AI in NEOM city. Energy AI
2021, 3, 100038. [CrossRef]
7. Raju Pendem, S.; Mikkili, S.; Rangarajan, S.S.; Avv, S.; Collins, R.E.; Senjyu, T. Optimal hybrid PV array topologies to maximize
the power output by reducing the effect of non-uniform operating conditions. Electronics 2021, 10, 3014. [CrossRef]
8. Schleifer, A.H.; Murphy, C.A.; Cole, W.J.; Denholm, P.L. The evolving energy and capacity values of utility-scale PV-plus-battery
hybrid system architectures. Adv. Appl. Energy 2021, 2, 100015. [CrossRef]
9. Hafeez, G.; Imran Khan, I.; Jan, J.; Shah, I.A.; Farrukh, A.F.; Derhab, A. A novel hybrid load forecasting framework with intelligent
feature engineering and optimization algorithm in smart grid. Appl. Energy 2021, 299, 117178. [CrossRef]
10. Mukhatov, A.; Thao, N.G.M.; Do, T.D. Linear Quadratic Regulator and Fuzzy Control for Grid-Connected Photovoltaic Systems.
Energies 2022, 15, 1286. [CrossRef]
11. Pande, J.; Nasikkar, P.; Kotecha, K.; Varadarajan, V. A review of maximum power point tracking algorithms for wind energy
conversion systems. J. Mar. Sci. Eng. 2021, 9, 1187. [CrossRef]
12. Verma, P.; Alam, A.; Sarwar, A.; Tariq, M.; Vahedi, H.; Gupta, D.; Shah Noor Mohamed, A. Meta-heuristic optimization techniques
used for maximum power point tracking in solar pv system. Electronics 2022, 10, 2419. [CrossRef]
Electronics 2023, 12, 592 25 of 26

13. González-Castaño, C.; Restrepo, C.; Kouro, S.; Rodriguez, J. MPPT algorithm based on artificial bee colony for PV system. IEEE
Access 2021, 9, 43121–43133. [CrossRef]
14. Ortiz Valencia, P.A.; Ramos-Paja, C.A. Sliding-mode controller for maximum power point tracking in grid-connected photovoltaic
systems. Energies 2015, 8, 12363–12387. [CrossRef]
15. Abbes, H.; Abid, H.; Loukil, K.; Toumi, A.; Abid, M. Etude comparative de cinq algorithmes de commande MPPT pour un
système photovoltaïque. J. Renew. Energ. 2014, 17, 435–445.
16. Marinić-Kragić, I.; Nižetić, S.; Grubišić-Čabo, F.; Papadopoulos, A.M. Analysis of flow separation effect in the case of the
free-standing photovoltaic panel exposed to various operating conditions. J. Clean. Prod. 2018, 174, 53–64. [CrossRef]
17. Gaur, P.; Verma, Y.P.; Singh, P. Maximum power point tracking algorithms for photovoltaic applications: A comparative study.
In Proceedings of the 2nd International Conference on Recent Advances in Engineering & Computational Sciences (RAECS),
Chandigarh, India, 21–22 December 2015; pp. 1–5. [CrossRef]
18. Enany, M.A.; Farahat, M.A.; Nasr, A. Modeling and evaluation of main maximum power point tracking algorithms for photo-
voltaics systems. Renew. Sustain. Energy Rev. 2016, 58, 1578–1586. [CrossRef]
19. Bhatnagar, P.; Nema, R.K. Maximum power point tracking control techniques: State-of-the-art in photovoltaic applications. Renew.
Sustain. Energy Rev. 2013, 23, 224–241. [CrossRef]
20. Nelatury, S.R.; Gray, R. A maximum power point tracking algorithm for photovoltaic applications. In Proceedings of the Energy
Harvesting and Storage: Materials, Devices, and Applications IV, Baltimore, MD, USA, 28 May 2013. [CrossRef]
21. Mohamed, S.A.; Abd El Sattar, M. A comparative study of P&O and INC maximum power point tracking techniques for
grid-connected PV systems. SN Appl. Sci. 2019, 1, 174. [CrossRef]
22. García, E.; Ponluisa, N.; Quiles, E.; Zotovic-Stanisic, R.; Gutiérrez, S.C. Solar panels string predictive and parametric fault
diagnosis using low-cost sensors. Sensors 2022, 22, 332. [CrossRef]
23. Abid, A.J.; Al-Naima, F.M. A Photovoltaic Measurement System for Performance Evaluation and Faults Detection at the Field.
Int. J. Autom. Smart Technol. 2020, 10, 409–420. [CrossRef]
24. Yilmaz, U.; Kircay, A.; Borekci, S. PV system fuzzy logic MPPT method and PI control as a charge controller. Renew. Sustain.
Energy Rev. 2018, 81, 994–1001. [CrossRef]
25. Seguel, J.L.; Seleme, S.I., Jr.; Morais, L.M. Comparative Study of Buck-Boost, SEPIC, Cuk and Zeta DC-DC Converters Using
Different MPPT Methods for Photovoltaic Applications. Energies 2022, 15, 7936. [CrossRef]
26. Rahman, C.M.; Rashid, T.A. A new evolutionary algorithm: Learner performance based behavior algorithm. Egypt. Inform. J.
2021, 22, 213–223. [CrossRef]
27. Kumar, A.; Bawa, S. A comparative review of meta-heuristic approaches to optimize the SLA violation costs for dynamic
execution of cloud services. Soft Comput. 2020, 24, 3909–3922. [CrossRef]
28. Jlidi, M.; Hamidi, F.; Abdelkrim, M.N.; Jerbi, H.; Abbassi, R.; Kchaou, M. Synthesis of an Advanced Maximum Power Point
Tracking Method for a Photovoltaic System: A Chaotic Jaya Logistic Approach. In Proceedings of the 4th International Conference
on Applied Automation and Industrial Diagnostics (ICAAID), Hail, Saudi Arabia, 29–31 March 2022. [CrossRef]
29. Rao, R. Jaya: A simple and new optimization algorithm for solving constrained and unconstrained optimization problems. Int. J.
Ind. Eng. Comput. 2016, 7, 19–34. [CrossRef]
30. Alghamdi, A.S. A Hybrid Firefly–JAYA Algorithm for the Optimal Power Flow Problem Considering Wind and Solar Power
Generations. Appl. Sci. 2022, 12, 7193. [CrossRef]
31. Zitar, R.A.; Al-Betar, M.A.; Awadallah, M.A.; Doush, I.A.; Assaleh, K. An intensive and comprehensive overview of JAYA
algorithm, its versions and applications. Arch. Comput. Methods Eng. 2022, 29, 763–792. [CrossRef]
32. Mohammad, M.; Abbassi, R.; Jerbi, H.; Waly, A.F.; Abdalqadir, A.H.; Rezvani, A. A new MPPT design using variable step size
perturb and observe method for PV system under partially shaded conditions by modified shuffled frog leaping algorithm- SMC
controller. Sustain. Energy Technol. Assess. 2021, 45, 101056. [CrossRef]
33. Hamidi, F.; Olteanu, S.C.; Gliga, L.I. Gradient Optimization Methods for Maximum Power Point Tracking in Photovoltaic Panels.
In Proceedings of the 15th European Workshop on Advanced Control and Diagnosis, Online ISBN, 14 June 2022. [CrossRef]
34. Hamidi, F.; Olteanu, S.C.; Popescu, D.; Jerbi, H.; Dincă, I.; Ben Aoun, S.; Abbassi, R. Model Based Optimisation Algorithm for
Maximum Power Point Tracking in Photovoltaic Panels. Energies 2020, 13, 4798. [CrossRef]
35. Zhang, Y.; Lundblad, A.; Campana, P.E.; Yan, J. Comparative study of battery storage and hydrogen storage to increase
photovoltaic self-sufficiency in a residential building of Sweden. Energy Procedia 2016, 103, 268–273. [CrossRef]
36. Yue, M.; Lambert, H.; Pahon, E.; Roche, R.; Jemei, S.; Hissel, D. Hydrogen energy systems: A critical review of technologies,
applications, trends and challenges. Renew. Sustain. Energy Rev. 2021, 146, 111180. [CrossRef]
37. Petkov, I.; Gabrielli, P. Power-to-hydrogen as seasonal energy storage: An uncertainty analysis for optimal design of low-carbon
multi-energy systems. Appl. Energy 2020, 274, 115197. [CrossRef]
38. Lei, Q.; Wang, B.; Wang, P.; Liu, S. Hydrogen generation with acid/alkaline amphoteric water electrolysis. J. Energy Chem. 2019,
38, 162–169. [CrossRef]
39. Escamilla-García, A.; Soto-Zarazúa, G.M.; Toledano-Ayala, M.; Rivas-Araiza, E.; Gastélum-Barrios, A. Applications of artificial
neural networks in greenhouse technology and overview for smart agriculture development. Appl. Sci. 2020, 10, 3835. [CrossRef]
40. Huang, J.C.; Ko, K.M.; Shu, M.H.; Hsu, B.M. Application and comparison of several machine learning algorithms and their
integration models in regression problems. Neural Comput. Appl. 2020, 32, 5461–5469. [CrossRef]
Electronics 2023, 12, 592 26 of 26

41. Available online: https://www.meteocontrol.com/fileadmin/Daten/Dokumente/ES/1_Photovoltaik_Monitoring/Accesorios/

Sensores/Irradiaci%C3%B3n/Sensores_de_radiaci%C3%B3n_solar_de_silicio/DB_Irradiance_sensor_Si-Series_en.pdf (accessed
on 12 November 2022).
42. Sajjad, U.; Hussain, I.; Raza, W.; Sultan, M.; Alarifi, I.M.; Wang, C.-C. On the Critical Heat Flux Assessment of Micro- and
Nanoscale Roughened Surfaces. Nanomaterials 2022, 12, 3256. [CrossRef]
43. Gaspar, A.; Oliva, D.; Cuevas, E.; Zaldívar, D.; Pérez, M.; Pajares, G. Hyperparameter Optimization in a Convolutional Neural
Network Using Metaheuristic Algorithms. In Metaheuristics in Machine Learning: Theory and Applications; Oliva, D., Houssein,
E.H., Hinojosa, S., Eds.; Studies in Computational Intelligence; Springer: Cham, Switzerlands, 2021; Volume 967. [CrossRef]
44. Sajjad, U.; Hussain, I.; Hamid, K.; Muhammad, A.H.; Wang, W.; Yan, W. Liquid-to-vapor phase change heat transfer evaluation
and parameter sensitivity analysis of nanoporous surface coatings. Int. J. Heat Mass Transf. 2022, 94, 123088. [CrossRef]
45. Chen, Z.; Francis, A.; Li, S.; Liao, B.; Xiao, D.; Ha, T.T.; Cao, X. Egret Swarm Optimization Algorithm: An Evolutionary
Computation Approach for Model Free Optimization. Biomimetics 2022, 7, 144. Available online: https://sciprofiles.com/profile/
1993400 (accessed on 2 November 2022). [CrossRef]
46. Hamidi, F.; Aloui, M.; Jerbi, H.; Kchaou, M.; Abbassi, R.; Popescu, D.; Dimon, C. Chaotic particle swarm optimisation for
enlarging the domain of attraction of polynomial nonlinear systems. Electronics 2020, 9, 1704. [CrossRef]
47. Rabeh, A.; Abdelkader, A.; Ali, A.H.; Seyedali, M. An efficient salp swarm-inspired algorithm for parameters identification of
photovoltaic cell models. Energy Convers. Manag. 2019, 179, 362–372. [CrossRef]
48. Zeb, K.; Islam, S.U.; Din, W.U.; Khan, I.; Ishfaq, M.; Busarello, T.D.C.; Kim, H.J. Design of fuzzy-PI and fuzzy-sliding mode
controllers for single-phase two-stages grid-connected transformerless photovoltaic inverter. Electronics 2019, 8, 520. [CrossRef]
49. Karami-Mollaee, A.; Barambones, O. Dynamic Sliding Mode Control of DC-DC Converter to Extract the Maximum Power of
Photovoltaic System Using Dual Sliding Observer. Electronics 2022, 11, 2506. [CrossRef]
50. Velasco, J.; Calvo, I.; Barambones, O.; Venegas, P.; Napole, C. Experimental validation of a sliding mode control for a stewart
platform used in aerospace inspection applications. Mathematics 2020, 8, 2051. [CrossRef]
51. Kabilan, R.; Chandran, V.; Yogapriya, J.; Karthick, A.; Gandhi, P.P.; Mohanavel, V.; Manoharan, S. Short-term power prediction of
building integrated photovoltaic (BIPV) system based on machine learning algorithms. Int. J. Photoenergy 2021, 2021, 5582418.
[CrossRef]
52. Rai, A.; Shrivastava, A.; Jana, K.C. A CNN-BiLSTM based deep learning model for mid-term solar radiation prediction. Int. Trans.
Electr. Energy Syst. 2021, 31, e12664. [CrossRef]
53. Kaba, K.; Sarıgül, M.; Avcı, M.; Kandırmaz, H.M. Estimation of daily global solar radiation using deep learning model. Energy
2018, 162, 126–135. [CrossRef]
54. Olatomiwa, L.; Mekhilef, S.; Shamshirband, S.; Petković, D. Adaptive neuro-fuzzy approach for solar radiation prediction in
Nigeria. Renew. Sustain. Energy Rev. 2015, 51, 1784–1791. [CrossRef]
55. Khosravi, A.; Nunes, R.O.; Assad, M.E.H.; Machado, L. Comparison of artificial intelligence methods in estimation of daily global
solar radiation. J. Clean. Prod. 2018, 194, 342–358. [CrossRef]
56. Halabi, L.M.; Mekhilef, S.; Hossain, M. Performance evaluation of hybrid adaptive neuro-fuzzy inference system models for
predicting monthly global solar radiation. Appl. Energy 2018, 213, 247–261. [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.