Hindawi

International Transactions on Electrical Energy Systems

Volume 2022, Article ID 1278492, 15 pages
https://doi.org/10.1155/2022/1278492

Research Article
Optimization and Performance Improvement of Grid-Connected
PV Plant Based on ANN-PSO and P&O Algorithms

1 2
Abdalftah Hamed Ali and Atabak Najafi
1
Omdurman Islamic University, Faculty of Engineering Science, Omdurman, Sudan
2
Eskisehir Osmangazi University, Institute of Science and Technology, Odunpazarı, Turkey

Correspondence should be addressed to Abdalftah Hamed Ali; abdalftahhamedali08@gmail.com

Received 11 June 2022; Revised 2 September 2022; Accepted 5 October 2022; Published 20 October 2022

Academic Editor: Ci Wei Gao

Copyright © 2022 Abdalftah Hamed Ali and Atabak Najafi. This is an open access article distributed under the Creative Commons
Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is
properly cited.
This research investigated the performance of a 5 MW PV grid-connected plant in Al Fashir City, Sudan. The research aims to
improve the performance and increase the efficiency of the Al Fashir plant by identifying the maximum power point and
increasing the tracking efficiency based on the algorithms developed. The PV systems benefit from MPPT approaches because they
improve power output and energy delivery to the load while also extending the useful life of the PV system. The P&O algorithm
performance is compared to the ANN trained by the PSO method by a set of solar radiation values. However, time response,
oscillation, and stability are the three most important factors to consider when evaluating the effectiveness of any MPPTalgorithm.
The results show that the ANN trained by the performance of the PSO algorithm was better in time response, tracking speed, and
oscillation than the P&O algorithm and could identify the new power point quickly. The results of this study will assist in resizing
the PV plant and improve the operation performance and efficiency to provide affordable and reliable power accessible to the
people in Al Fashir city.

1. Introduction increased rapidly over the past decade due to the rising
demand for environmentally friendly sources. Despite the
One of the most recent innovations among the various forms delay caused by the COVID-19 epidemic in many supply
of renewable energy used today is solar energy. Moreover, it chains, according to the International Renewable Energy
is anticipated that renewable energy sources such as solar Agency (IRENA), renewable energy capacity increased
energy will soon replace traditional energy sources for many globally to at least 2537 GW at the beginning of 2020,
reasons, such as depletion, rising costs, and health and representing a 50% increase over 2019. However, solar
environmental concerns. However, increasing renewable energy increased by 24% to 94 GW during the same year,
energy sources has many benefits, including reducing while wind energy increased by 18% to 111 GW [1].
greenhouse gas emissions and air pollution. Furthermore, As the capacity of photovoltaic systems increases sub-
renewable energy provides more than half of the world’s stantially, the influence of PV modules on the electricity grid
electrical needs, reducing fossil fuel dependency by more cannot be overlooked. It could produce obstacles to the grid,
than one-third. Solar System prices have started to decrease such as flickering, harmonics, and deterioration of the grid’s
due to advancements in manufacturing, performance, and stability. In order to enhance photovoltaic system perfor-
quality of devices. With the advancement of solar technology mance and maintain power stability, it is crucial to meet the
and the widespread adoption of solar systems, the cost of technical requirements of the photovoltaic system, such as
solar energy is expected to fall significantly. fault-ride-through capabilities and harmonic distortion
The photovoltaic system is one of the most widely used control. However, the stability during normal and fault
solar energy sources. The popularity of photovoltaics has situations should be investigated when photovoltaic
2 International Transactions on Electrical Energy Systems

modules interlink the network. However, the author [2] has consistency, and ability to distinguish between global and
studied the penetration of different renewable energy local maximum power points. The author [10] proposed
sources to determine how to evaluate their operational two very efficient MPPT controllers using innovative slime
characteristics and efficiency. Consequently, the effective- mould optimization (SMO) and an improved salp swarm
ness of the systems was evaluated using technical, economic, optimization technique (ISSA) to change temperature and
and environmental restrictions and comparative cost anal- irradiance. The suggested controllers are compared to
ysis of various operational scenarios. Additionally, the conventional and recent algorithms. The results demon-
sensitivity analysis for fuel, PV, battery pricing, and load strated that the suggested controllers had the best transient
profile was investigated to depict the impacts of altering and steady-state performances. Also, the author [11] pre-
significant parameters on system performance. sented the MPPT approach based on the dragonfly opti-
Generally, the performance of solar cells is affected by mization algorithm (DOA) for optimizing the photovoltaic
climatic variables (sunlight and temperature) and load system. The overall performance of the suggested approach
impedance. The fundamental drawback of photovoltaic was examined and compared to rigorous techniques such
systems is the poor energy conversion performance (in low as IPSO, PSO, PO, and other algorithms. The simulation
radiation and temperature conditions). Maximum power results showed that the proposed DOA approach beats
point tracker (MPPT) is used for optimal efficiency to other techniques in terms of response time, oscillations
improve systems performance. Nevertheless, the PV system reduction, resilience, speed of convergence of accuracy, and
uses MPPT approaches to improve power output and extend power efficiency.
the whole Solar System lifespan. Another study [4] introduced three metaheuristic op-
There was a focus on photovoltaic modules as a source timization approaches to tune the fractional-order incre-
of losses caused by low irradiance or high temperature. mental conductance FO parameters to control the output
However, it is recommended to use the MPPT algorithm by voltage of the Solar System, which are PSO, ant colony
keeping the output voltage for the module constant at a DC optimization, and AntLion Optimizer. Furthermore, the
level and mitigating harmonics signals as a solution. author [12] designed the MPPT method that drives a mix of
Meanwhile, researchers often divide these methods into the ANN and PSO to locate the global peak point. The study
two broad categories, namely, traditional techniques, such used MATLAB and hardware simulations to compare the
as perturbation and observation (PO) and hill-climbing ANN with the performance of the traditional swarm algo-
(HC). Additionally, there are modern techniques such as rithm, producing comparable results. The author of [13]
fuzzy logic, metaheuristic techniques (genetic algorithms, presents a comparative study for monitoring the highest
particle swarm optimization, and ant colony), and artificial power point of a neural network using the ANN-MPPT
neural networks [3, 4]. Compared to modern techniques, technique, PO, and incremental conductance MPPT. The
traditional algorithms have limitations such as a delay in simulation results in MATLAB show that the suggested
reaching the maximum power point (MPP), oscillations, ANN method improves efficiency and reduces oscillations
getting trapped at the local MPP, and the inability to compared to the PO and IC techniques. In addition, to
identify global MPP. In recent years, approaches have been increase the performance of the PV system and stabilize the
increasing that make metaheuristic MPPT algorithms MPPT, the author [14] presented a hybrid adaptive con-
better than the traditional algorithm [5]. For example, the troller based on the sine-cosine algorithm (SCA) and its
author [6] developed an algorithm to calculate the voltage excess efficiency over the invasive weed optimization
reference, which generates constant power from the PVPP. approach.
However, the desirable power is greater than the present After a comprehensive analysis, it was observed that
power, and the algorithm also demonstrates that the de- most of the research used MPPT approaches to combat the
sired power reference is more than the maximum available inefficiency of solar power facilities. However, no authors
power of the PVPP. On the other hand, the study in [7] compare the daily performance of real data with the PSO-
proposed a new method for the photovoltaic system that and PO-trained ANN algorithm. The comprehensive
depends on an adaptive prediction algorithm to regulate analysis was presented based on the proposed techniques
the output power of panels and eliminate the total har- to investigate the performance of a 5 MW PV grid-con-
monics distortion. nected plant in Al Fashir, Sudan. By comparing the plant’s
HOMER Pro and RETScreen simulators have been used precise operation to the results of suggested algorithms, a
in [8] as the performance predictors for large-scale solar neural network was trained by PSO and PO on the
power plants of 20 MW in a hot environment using real MATLAB platform. The objective of the research is to
operating data. Hence, the findings revealed that the per- enhance the efficiency of the solar Al Fashir plant by
formance of photovoltaic power plants is affected by cell increasing the tracking response and consequently in-
technology, insolation, and environmental factors, partic- creasing the tracking efficiency by using developed al-
ularly temperature. gorithms for identifying the MPP. This study used ANN-
Meanwhile, the author [9] examined the performance PSO for a sizeable single-stage plant and compared it to
of improved PSO compared to conventional PSO and the PO algorithm. The strategies offered aim to improve
MPPT algorithms under partly shade circumstances using the plant’s performance, which can be considered inno-
MATLAB/Simulink. The simulation outcome indicated vative for this geographic region. The following are the
improved PSO’s quick tracking speed, exceptional major contribution of the study:
International Transactions on Electrical Energy Systems 3

(i) Two efficient and improved algorithms have been generated by the photovoltaic array to alternating current
investigated in the MATLAB environment electricity. Over the previous several decades, various al-
(ii) The recommended methodologies are implemented ternative topologies have been created and deployed [18, 25].
successfully in the PV system and compared to the As mentioned previously, the multistring architecture was
actual output perhaps the most prevalent form of a photovoltaic system.
The real benefits of this topology are that it is self-contained,
(iii) A comparison was made between the proposed
provides better protection for power sources, is more secure
PSO-trained ANN and the P&O algorithm
during installation and maintenance, and is inexpensive.
Figure 3 shows how PV modules are tied to the inverter as a
2. Grid-Connected Components multistrand topology for only one inverter. The power plant
Grid-connected systems are designed to convert as much has 84 and 60 kW DC/AC converters. The input range of
irradiant energy as feasible into actual energy. Research 570–800 VDC assures the steadiness of AC output voltage.
studies have faced challenges in single-phase and three- However, there are 20 strings, and every row has 20 modules
phase systems such as maximum power consumption, rapid (20 ∗ 20) to generate 1500 Mw; every 25 inverters are
and regular power fluctuations, effects on network quality, coupled to a 1500 KVA transformer (0.4/33 KV) and then
current return, and mismatch between PV output and connected to the grid.
network demand [15]. The GCVS system consists of a matrix
of series/parallel photovoltaic modules, which turn the light 3. Maximum Power Point Tracking (MPPT)
from the sun to the current and a power conditioning device
that transforms DC into AC, as shown in Figure 1. The It is important to note that solar panels are nonlinear, and
power generated is delivered to the network and used by the the characteristics of these devices are highly dependent on
load demand. In certain circumstances, storage devices are the amount of light and temperature available [21].
employed to increase the availability of the power produced Meanwhile, the technical efficiency factor of an accessible
by the photovoltaic system. Here, we are trying to simulate photovoltaic panel is approximately 20%. A real-time MPPT
the Al Fashir station exclusively, so in the following sections, is critical to maximizing the energy harvested by photo-
additional specific information regarding the various voltaic panels; researchers have developed different MPPT
components of the PV system is provided. techniques. The most prevalent strategies are PO and the IC
algorithm. However, choosing the optimal MPPT approach
to utilize remains a challenge. As a result, more research is
2.1. PV Modules. A photovoltaic module is a collection of cells being conducted on MPPT approaches to develop a system
coupled in series and parallel to generate the required capacity, with good performance in cost savings, simplicity of in-
and a photovoltaic array is a collection of modules connected in stallation, sensing accuracy, and adaptation to diverse
parallel or series to provide the required power [16]. Generally, photovoltaics [22]. Partial shadowing of PV arrays is one of
the series combination of solar cells forms a PV module the critical issues faced by MPPT. MPPT is used in DC/AC
consisting of a single diode with series resistance and shunt inverters to maximize the efficiency of PV modules and to
resistance. The current source is the photon-generated current determine the MPP for various characteristics under any
(Iph) produced by photovoltaic (PV) proportional to sun ir- environmental circumstance. We used two algorithms,
radiation. Resistance (Rs) indicates contacts and connection- namely, PO and ANN, trained by the PSO. Those algorithms
related losses, while parallel resistance represents the diode provide the reference voltage to the DC link controller under
leakage currents (Rsh) Figure 2. Also, the mathematical different conditions.
equation used to calculate the I–V of a solar cell quantitatively
is given in equation (1) [17]. A proper solar cell device must
account for the effects of the series and parallel resistances to 3.1. PO Algorithm. Because of its simplicity, the PO ap-
function correctly. The electrical properties of a photovoltaic proach is the most utilized MPPT technique in photovoltaic
module are often represented as the relationship between the systems. This approach ensures consistent accuracy, doubles
cell voltage and current and the relationship between the cell tracking speed, and transforms significant energy into
voltage and the output power, among other things. power. PO is utilized as the maximum power tracking ap-
The circuit gives the following relationship and the proach to manage voltage, current, and power across the
characteristic equation: solar module, significantly lowering the THD of the grid-
Iph � ID + IRsh + i, (1) connected photovoltaic system [3]. Using the PO method to
detect power fluctuations is possible by creating a pertur-
bation, adjusting the voltage, and detecting the resulting
V + RS V + Rs ∗ I power fluctuation. First and foremost, to achieve this goal, it
I � IPV − I0 􏼢exp􏼠 􏼡􏼣 − . (2)
Vt a RSH is necessary to measure the voltage (Vpv) and current (Ipv) of
the solar module under constant temperature and irradiance
conditions. These measurements are taken at time tk and
2.2. Inverters. Grid-connected photovoltaic systems connect compared to the previously recorded amounts taken at time
the photovoltaic array to the power grid through a power tk1 to determine if any differences exist. At tk, it is possible to
electronic inverter, which converts the DC electricity estimate the power (Ppv), which can then be compared to the
4 International Transactions on Electrical Energy Systems

+
Power Vac
Vdc Utility Grid
− Conditioning Unit

Vac
PV Unit

Local Loads

Figure 1: PV grid-connected block.

RS i

Iph D RSH v
iD iRsh

Figure 2: Electrical circuit of the PV cell.

Modul 1 Module 20
2x1x4 mm2
String 1
DCCB 01.01.03

2x1x4 mm2
Inverter: String 2
2

SMA STP 60 2x1x4 mm2

String 3
3

2x1x4 mm2
String 4
4

2x1x4 mm2
String 5
5

2x1x4 mm2
String 6
2P
6
21

2x1x4 mm2
315 A String 7
7

2x1x4 mm2
String 8
8

2x1x4 mm2
String 9
9

2x1x4 mm2
DC-disconnector String 10
INV 01.01.03
11 10
12

Figure 3: Multistring topology arrays coupled to one inverter.

power calculated at tk1. A positive change in power leads the systems, like photovoltaic panels. Like the central nervous
system’s operational voltage to shift in the same direction as system, it comprises essential and highly linked artificial
the increase, and when the change is negative, the opera- neurons; we can distinguish between single-layer net-
tional voltage shift in Figure 4 shows the flow chart. works of a single layer of neurons, and the information
can flow in any direction. Also, a multilayer is unidi-
rectional in which information only flows in one way and
3.2. ANN Algorithm. Since the 1970s, the ANN has been is made up of many nodes in more than three layers,
regarded as one of the most extraordinary possibilities for namely, input, hidden, and output, as can be seen in
computing systems. New neural network-based control- Figure 5 [23]. However, thousands of weighted links link
lers are ideal for this purpose for complex and inadequate these neurons, allowing messages to travel between them
issues where traditional approaches have not attained the easily. Countless messages arrive at each neuron, but only
requisite quickness, precision, or efficiency. Although one leaf. Additionally, the neurons in each layer are linked
modeling an application system using the ANN does not by their weights and bias terms in the preceding levels
require much expertise, precise data is needed to forecast (Figure 5). Equation (3) shows the weights computation
output functions as closely as possible to reality [22]. calculation [22].
There are two kinds of ANN architecture feedback net- n
works and feedforward, which are often used since they y � 􏽘 wij xj + bj . (3)
consume lower capacity during the execution stage and i�1
are very effective when used in conjunction with nonlinear
International Transactions on Electrical Energy Systems 5

Start

Measure V(k)&I(k)

P(k)−P(k− Yes
1)=0

NO

No P(k)−P(k− Yes
1)=0

NO V(k)−V(k− NO V(k)−V(k− YES

YES
1)>0 1)>0

Decrease Vref Increase Vref Decrease Vref Increase Vref

Return

Figure 4: P&O algorithm flow chart.

Hidden
layer
Input Hidden Output
layer
Output Input Output
T b3 layer Vmpp W W
b1
+ +
G
b4 b6
b b
b2 2 1

10 1
b5

Figure 5: Structure and graphical diagram of ANN.

The backpropagation (BP) technique is often used to inputs should be independent variables that have a mean-
improve ANN performance by altering node weights and ingful effect on the ANN response. Third, the architectural
bias terms, and the cost function calculation is generally choice, one of the most common forms of ANN, is the
selected as MSE and given by multilayered feedforward backpropagation, which is utilized
in a wide variety of applications and divided into input
1 n m 2 layers, a hidden layer containing neurons, and an output
MSE � 􏽘 􏽘 􏽨Y (i) − Tj (i)􏽩 . (4)
n i�1 j�1 j layer. Then, training and testing ANN is finding the weights
that comprise the ANN critical components. The training
Generating an ANN block is the first process in gen- procedure determines which weights minimize some mea-
eration of training patterns, which entails assembling a sure of overall error, such as the sum of squared errors (SSE)
collection of training patterns that encompasses all potential or mean squared errors (MSE). Accounting for the optimal
operational scenarios. Second, input selection for ANN number of hidden layers and units inside each is critical for
6 International Transactions on Electrical Energy Systems

feedforward ANN construction and addressing the trade-off Start

between computing time and best-fitting regression of the
ANN node. When an ANN has excessive units in the hidden Set PSO parameters
Vmax,Xmin,Xmax,C1,C2,
layers, the computing time comes to be excessive due to an W1,W2
overfitting regression.
Intinalize X,V

3.3. PSO Algorithm. PSO is a specialist field of machine

learning for solving large amounts of data and finding suitable Evaluate Fitness
function
solutions quickly. The PSO algorithm’s movement is inspired
by birds’ flocking behavior, which depends on the PSO opti-
Assign Initial Pbest
mizer’s individual and neighboring experiences at each particle & Gbest
step. The basic concept behind this approach is to identify an
optimized region where each space has the probability of
Unpdate X,V
containing a potential solution. The PSO algorithm comprises
four stages [22]. The PSO optimizer begins by searching by a
random particle value. This particle value is chosen according K=k+1
Evaluate Fitness
function
to the degree of the potential of solution spaces for various
optimizations. The previous and next best fitness values are
compared, and optimal solutions in the same space in the Unpdate Pbest & Gbest
second phase are searched. The best global positions are
NO K=
compared in the third stage to obtain the global fitness value. Maxite
These locations are computationally modified and maintained YES
for the following step in this phase, as stated by [23] Stopt

Vk+1
i � W ∗ Vki + r1 ∗ c1 ∗ 􏼐Pbi − XK
i 􏼑 Figure 6: PSO algorithm flow chart.
(5)
+ r2 ∗ c2 ∗ 􏼐Gbi − Xki 􏼑,
for the fundamental electrical frequency. Meanwhile, the
Xk+1 � Xki + Vki . (6) control circuit acts as the brain for the whole system [25, 26].
i

The fourth step determines and saves the optimal particle

in fitness evaluation to improve the particle movement steps 4.2. Implementation of Control Circuits. The MATLAB de-
in each iteration. These steps are repeated until a stopping sign approach that developed and modulated the proposed
criterion is met or the specified number of repetitions has grid-connected photovoltaic system is shown in Figure 8.
been reached, as can be seen in Figure 6, which shows the Photovoltaic arrays are coupled to the inverter through the
PSO procedure of PSO [24]. The system’s required accuracy DC link, measurement circuits, and the inverter control
and processing time suggest the stopping condition and circuit. The control system consists of five primary Simulink-
number of iterations. The primary disadvantages of PSO and based subsystems. Figures 9 and 10 demonstrate the MPPT
other metaheuristic optimizers are a lack of a robust here, and we used PO or ANN trained by the PSO algorithm
mathematical foundation and an inability to guarantee in this study. Since the current reference value changes
global optimum solutions [23] theoretically. The predictive according to the MPPT techniques, the inverter’s output
function of the ANN model is optimized using the PSO current fluctuates in response to the light intensity trajectory.
method. The optimization process has two stages, namely, When the MPPTsystem is activated, the VDC reference signal
determining the optimal feedforward topology and selecting of the inverter VDC regulator is automatically adjusted to get
the initial weights for the ANN model. a DC voltage that will extract the most significant amount of
power from the PV array. Also, other VDC regulator circuits
4. Methodology determine the Id reference current for the current regulator
using the voltage-dependent current regulator. Current reg-
4.1. Main Block Diagram. The grid systems’ essential com- ulator circuit determines the reference voltages Vd for the
ponents are a photovoltaic array, DC link, MPPT, inverter, inverter based on the current references Id and Iq. A phase
filter, and transformer coupling to a utility, as can be seen in lock loop (PLL) control circuit is required for synchronization
Figure 7. The capacitor in the DC connection is charged by and adjusts the internal frequency oscillator and tracks the
the direct current electricity supplied by the photovoltaic frequency and phase of a sinusoidal signal [21]. However, the
array. The inverter converts DC to AC at a sinusoidal voltage inverter’s output currents IABC are acquired. After applying
and frequency comparable to the power grid. The trans- Park’s transformation through PI controllers, Id and Iq are
former raises the inverter voltage to the nominal value of the generated and utilized for the system’s active and reactive
grid while also providing electrical isolation between the power regulation. Then, PWM block serves as the switching
photovoltaic system and the power distribution network. pulses and generates signals for the inverter depending on the
The harmonic filter cancels all harmonic components except reference voltages. The grid voltages establish the phase angle
International Transactions on Electrical Energy Systems 7

Control unit
MPPT
Controller control

DC-buses Filter Transformer

Power electronics
Utility Grid
PV Arrays Multilevel converter Optional Stage

Figure 7: Block diagram of the on-grid photovoltaic system (source [9]).

MPPT_On
1

[m_inv] <Vabc>
[m_inv] Grid
<Iabc>
m 1
<V_PV> [m_solar] PV m
<I_PV>
Measures

m [m_solar] Inverter Control

1 Ir
Ir [m_inv]
+
1
A
[45] 2
2 T
− B
T 3
DC Link Grid Inverter C
JA Solar JAP72S01-325/SC
20-module string
811 parallel strings

Figure 8: The photovoltaic system circuit.

wt wt
2 Vabc_prim
Uref
Vabc_prim 1
VdVq_prim VdVq_meas Uabc_ref Vabc
z g 1
Param Param

3 Iabc_prim
IdIq_prim IdIq_meas VdVq_conv VdVq_conv 1 vdc 0
Enabled Iabc_prim
D PLL & Measurements Display
ModWave
IdIq_ref vneutral
0
5 V PandO 0 Uabc_ref Generation
max (m) = 1
V_PV Current Regulator PWM Generator
Id_ref
4 Vdc_meas (Three-phase, Three-level)
6 I
Vdc_meas
I_PV Id_ref

Vdc_ref
On

1 Vnom_dc
VDC Regulator

Figure 9: The PO proposed circuit.

for a three-phase proportional-integral (PI) regulated PLL. system. The number of strings is 811, each consisting of 20
Meanwhile, the inverter output will be frequency and phase series modules. The type of module used is polycrystalline
angle synchronized with the grid voltage. from JA Solar manufacturer 325 W, and the efficiency for
each module is 16.73%. Table 1 shows all the PV module I–V
specifications under different weather conditions.
4.3. Photovoltaic Array Modeling. The system essentially
consists a matrix of series/parallel photovoltaic modules
from the block diagrams, which convert light from the sun 4.4. MPPT Proposed Algorithms. As explained previously,
into current (dc) electricity. The number of modules used for photovoltaics is characterized by poor efficiency and non-
that plant design is 16220 with 325 W from solar in the PV linear PV system properties with a single MPP. Thus,
8 International Transactions on Electrical Energy Systems

wt wt
2 Vabc_prim
Uref
Vabc_prim 1
VdVq_prim VdVq_meas Uabc_ref Vabc
z g 1

3 Iabc_prim IdIq_prim 1
IdIq_meas VdVq_conv VdVq_conv vdc
Iabc_prim
PLL & Measurements ModWave
G IdIq_ref vneutral
0 Uabc_ref Generation
0
T Input NNET Output max (m) = 1
Id_ref Current Regulator PWM Generator
4 Vdc_meas
(Three-phase, Three-level)
Vdc_meas Id_ref
Vdc_ref

Feed-Forward Neural Network VDC Regulator

On
1

Vnom_dc

Figure 10: ANN trained with PSO algorithm control circuit.

Table 1: Technical specifications of the 325 W PV module from JA topology and optimizing the model’s initial weight values.
Solar company. Running this hybrid algorithm yields the optimal starting
Rated maximum power 325 W weights, which are then used to optimize the rest of the
Open circuit voltage 46.38 V procedure. Then, the command “nntool” used in the pro-
Maximum power voltage 37.39 V gram to optimize starting weights is utilized for training the
Short circuit current 9.17 A ANN model. When we executed the collected M file, the
Maximum power current 8.69 A input data (irradiation, temperature), and the output voltage
Module efficiency 16.73% for different conditions, we obtained Figure 11. For ANN
Power tolerance −0∼+5 W training, there are many tools; one of the selected is the
input-output and curve fitting toolbox. However, the
monitoring the solar array’s maximum power point is a number of hidden neurons was ten as can be seen in Fig-
critical component of a photovoltaic system. The MPPT ure 5. Moreover, the Levenberg–Marquardt algorithm has
technique is based on a control system that keeps the PV been chosen, and the output is matched with trained data (as
system module operating at its maximum power point can be seen in Figure 12 plot regression and best cost
throughout the day. Here, we would compare the result of function for the iteration.
two algorithms, namely, PO and ANN, trained by PSO.

4.5. Three-Level NPC. PV inverter is a mature technology

4.4.1. P&O Algorithm. In the PO algorithm, the current and consisting of electronic power switches operated by the
voltage of a photovoltaic module are monitored at any given control strategy. The circuit used in the research was NPC
time and then the operating point of the photovoltaic switching devices with a sinusoidal PWM-controlled in-
module changes with the change in direction and reaches the verter (Figure 13). The three-phase PV inverter’s voltage
MPP when dp/dv � 0. We can then figure out the operating and current-controlled techniques are generally used in
point corresponding to the maximum power by observing applications. A voltage source inverter control capability
the voltage increments and decrements. The rise or fall in has key advantages, namely, lower switch voltage ratings, a
power figure below depicts the operation. Due to its ease of broad harmonic spectrum, and fast dynamic response. In
execution and cheap cost, PO control is a popular meth- this design, the neutral-point voltage must be controlled to
odology. Unfortunately, it is inefficient in precision since it avoid increasing the device voltage stress and output
increases and decreases the MPP [26]. Figure 9 shows the voltage. Because there are more devices in series with the
proposed PO circuit. current channel and losses might increase if the current
levels are the same, the 3-level inverter is more efficient
than the standard 2-level inverter. Therefore, when a
4.4.2. ANN Trained by PSO. In this study, PSO is used to considerable DC-bus voltage increase is conceivable, 3-
improve the training strategy of the ANN model by finding level (or multilayer) topologies are intriguing. The benefits
the suitable topology for the feedforward ANN model and of multilayer inverters in harmonic reduction apply, and
then finding the best way to set the initial weights. Specif- the cost decrease in the output filter partly offsets the cost
ically, G and T of the atmospheric conditions are employed increase due to the increased device count [26]. A DC-link
as inputs to the ANN block, with a voltage reference output voltage controller maintains a consistent voltage across the
resulting from the model, as can be seen in Figure 10. Then, capacitor by modulating the active current supplied into
the voltage goes to the VDC regulator to determine the Id the inverter. The minimum DC-Link voltage for the in-
reference current. The process begins with selecting ANN verter can be computed with equation (7) [27]. Two filter
International Transactions on Electrical Energy Systems 9

1200 850

1000 800

800 750

600 700

400 650

200 600

550
0 2000 4000 6000 8000 10000 0 2000 4000 6000 8000 10000

Figure 11: Possibility of radiation, temperature, and output voltage.

Training: R=1 Validation: R=1

Output ~= 1*Target + −0.0039

Output ~= 1*Target + 5.1e−05
750 750
748 748
101 746 746
744 744
742 742
740 740
738 738
100
740 745 750 740 745 750
Target Target
Data Y=T Data Y=T
Best Cost

Fit Fit
10−1 Test: R=1 All: R=1

Output ~= 1*Target + −0.00021

Output ~= 1*Target + 0.0022

750 750
748 748
746 746
10−2 744 744
742 742
740 740
738 738

10−3 740 745 750 740 745 750

0 20 40 60 80 100 Target Target

Iteration Data
Fit
Y=T Data
Fit
Y=T

Figure 12: The best cost function for the iteration and plot regression.

Uref
g 1
1 + m
Harmonic Filter
Inv+ BL Vabc 1
Iabc
2 N A A a a A
N
B B b b B 4
3 − 5 A
3-Level C C c cD1 Yg C
Inv−− 6 B
L B1 C
A

C
B

10 kvar

Figure 13: Three-level NPC.

types are utilized to smooth harmonics before coupling the overshoots [28]. The choke circuit is dominated by in-
inverter to the 260 V/25 KV transformer and then to the ductance, as shown by
grid. A series RL choke and a C-filter are parallel config- √�
VDC−LINL ≧2 2.V phase, (7)
urations used to reduce harmonics, simultaneously in-
creasing reactive power. The RL choke, sometimes referred
to as a line reactor, reduces harmonics generated by the 1 V kV2L−L
Lchoke � k ∗ 􏼠 􏼡 ∗ 􏼠√� L−L 􏼡 � . (8)
inverter and protects the equipment from transient 2πfac 3 Irms 2πfac Pac
10 International Transactions on Electrical Energy Systems

pvM

A A a

B B b

C C Yg Yg c

154MW-33 kV 33/0.4 kV Grid Connected G

MAIN GRID Solar PV Array
Discrete
1e−06 s.
T
voltage grid
current
Vabc2 Vabc P -K-
Iabc Q

A
B
C
to kW2
Iabc2

A
B
C
Power (kW)
P = 500e3 W
pvM <V_PV> x Load 1
-K-
<I_PV>
P = 500e3 W
to kW1

PV
Vcl1 Vabc P -K- Vpv
voltage
Iabc Q to kW7
Icl1
Ipv
current
Pv PCC
Vcl2 Vabc P -K-
Iabc Q to kW6
Icl2
Voltage
Current
voltage Load 1
current
Load 2

Figure 14: The main circuit diagram.

4.6. Grid Circuit. Figure 14 shows the system implemented the power and efficiency are close to the precise theoretical
in MATLAB software; three feeders and 1 MW DC represent values corresponding to the irradiation levels.
the local load, and a transmission line sets up the trans- Even for voltage and current curves for the same case, in
former-coupled DC system for the utility. Figure 16(a), we observed that the voltage curve for ANN-
PSO has stable performance and recorded a constant value
5. Simulation Results and current. On the other hand, the photovoltaic array’s
voltage for the PO algorithm from time zero to time equal
The research implemented using two cases to assess resilience one is not like the ANN since the value fluctuates between
of the proposed system built-in MATLAB, as shown in Fig- 510 at the beginning and then increases until 750 at the
ure 14, under rapidly changing atmospheric circumstances. endpoint. For the ANN, the photovoltaic voltage is stable,
First, we consider the following scenario: the temperature is and the output is changed between 600 at the beginning and
constant (T � 25°C) while varying the irradiation as a simple 840 V. Nevertheless, in the current curve, a fluctuation at the
case. Second, we adjust the temperature and irradiance si- beginning for ANN-PSO compared to the PO algorithm is
multaneously to simulate the real performance of the Al Fashir less. Despite the PO algorithm’s current values being more
station as a complex case. The performance of the PO algo- than ANN-PSO, the PO current curve is not like the ANN
rithm is compared to the ANN trained by the PSO method by a current curve, which is generally stable at a constant value.
set of solar radiation values. However, time response, oscil- Figure 17 shows the voltage and current output of the
lation, and stability are the three most important factors to three-phase inverters. Because voltage is one of the syn-
consider when evaluating the effectiveness of any MPPT al- chronization criteria that connect any power source to the
gorithm. Initially, we considered only two irradiation condi- grid, it must be more stable over time, even when the ir-
tions (1000 and 500 W/m2) as simple cases. We kept the radiance fluctuates. However, a three-phase current is
temperature constant at 250 C. Solar radiation starts at 1000 W/ depicted as Ia, Ib, and Ic from time T � 0, to time T �1 second.
m2. It rapidly drops to 500 W/m2 at 0.5 seconds to simulate the Because of the characteristic equation of the photovoltaic
transitory condition to determine how any methods respond to cell, the current varies as irradiance changes over time. On
the transient state. As we see in Figure 15(a), the ANN trained the other side, Figure 18(a) shows the voltage THD content
by the PSO method has stable tracking for power. In the case of for the ANN compared to the PO THD content in
1000 W/m2, the output power is greater than 5 MW compared Figure 18(b).
to the output power, which has not kept stable at a specific The data were collected randomly for two days (11 April
value and during the same condition has less value (<5 MW), as 2021 and 8 November 2021). The plant was simulated with
seen in Figure 15(b). nonuniform irradiance and temperature patterns on all solar
Meanwhile, when the radiation changes from 1000 W/ arrays. Radiation started rising from zero to a specific value
m2 to 500 W/m2, we can observe that the time response is (more than 1000 W/m2) and then reduced to zero
less for the ANN trained by PSO than for the PO algorithm. (Figures 19(a) and 19(b)). The waveforms of PV power
However, ANN-PSO slightly fluctuated before returning to a against simulation time were shown for all scenarios based
steady-state point, and PO fluctuated much more at this on solar irradiance patterns. The program’s execution time
state. This approach requires less time than PO to achieve was 15 seconds, but the number of hours from 6 am to 6 pm
tracking power from the transient to steady state. In all cases, is thirteen hours, each hour by one second with irradiation
International Transactions on Electrical Energy Systems 11

Solar PV power (kw) Solar PV power (kw)

5000 5000

4000 4000
Pv Power

3000 3000

2000 2000

1000 1000

0 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

to kW1
Irradiation
(a) (b)
Figure 15: Output power ANN trained by (a) PSO and (b) PO.

Solar PV Voltage Solar PV Voltage

850 750
800 700
750 650
700
600
650
550
600
500
Solar PV Current Solar PV Current
8000 8000
7000 7000
6000 6000
5000
5000
4000
4000
3000
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
(a) (b)

Figure 16: Voltage and current output. (a) ANN-PSO. (b) PO.

voltage
400 voltage:1
voltage:2
200 voltage:3
0
−200
−400

×104 current
current:1
1 current:2
current:3
0
−1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Figure 17: The phase voltages and currents at the inverter.

and temperature on a given day. For example, the first two suggested algorithms using the weather data for the days
second is six o’clock in the morning, and the radiation value determined. However, the following two graphs represent
is zero. The gradient continues until it reaches its maximum the G and T of those days; with different atmospheric
value in the middle of the day and then decreases to zero at conditions applied to the PO and ANN algorithms, it is more
six o’clock in the evening. Actual power production for those robust in delivering the optimal MPP under increasing and
two days in Figure 20 was compared with the findings of the decreasing radiation, as shown in Figure 19. The sun’s
12 International Transactions on Electrical Energy Systems

Fundamental (50 Hz) = 572.3, THD = 4.35% Fundamental (50 Hz) = 573.4, THD = 4.30%

Mag (% of Fundamental)
Mag (% of Fundamental) 2.5
2
2
1.5
1.5
1
1
0.5 0.5

0 0
0 100 200 300 400 500 600 700 800 900 1000 0 100 200 300 400 500 600 700 800 900 1000
Frequency (Hz) Frequency (Hz)

(a) (b)

Figure 18: THD for output voltage. (a) Using PO. (b) Using the ANN.

Temperature Curves
35
30
Temperature (oC)

25 Two-days Irradiation

Irradiation (W/m2)
20 2000
15 1500
10 1000
5 500
0 0 6:00 AM

8:00 AM

10:00 AM

12:00 PM

2:00 PM

4:00 PM

6:00 PM

6:00 PM
6:00 AM
8:00 AM
10:00 AM
12:00 PM
2:00 PM
4:00 PM
6:00 PM
6:00 PM

Hours per day Hours

11-Apr 11-Apr
8-Nov 8-Nov

(a) (b)

Figure 19: (a) Temperature curves for two days. (b) Solar irradiation.

Al-Fashir 5 MWp Power Plant: Monday, November 8, 2021 Al-Fashir 5 MWp Power Plant: Sunday, April 11, 2021
6000 6000

5000 5000

4000 4000
Power (kW)
Power (kW)

3000 3000

2000 2000

1000 1000

0 0
1:00 AM
2:00 AM
3:00 AM
4:00 AM
5:00 AM
6:00 AM
7:00 AM
8:00 AM
9:00 AM
10:00 AM
11:00 AM
12:00 PM
1:00 PM
2:00 PM
3:00 PM
4:00 PM
5:00 PM
6:00 PM
7:00 PM
8:00 PM
9:00 PM
10:00 PM
11:00 PM
12:00 AM

1:00 AM
2:00 AM
3:00 AM
4:00 AM
5:00 AM
6:00 AM
7:00 AM
8:00 AM
9:00 AM
10:00 AM
11:00 AM
12:00 PM
1:00 PM
2:00 PM
3:00 PM
4:00 PM
5:00 PM
6:00 PM
7:00 PM
8:00 PM
9:00 PM
10:00 PM
11:00 PM
12:00 AM

Power Power
(a) (b)

Figure 20: Plant recorded output power. (a) April. (b) 8 November.

irradiance begins to fluctuate from time zero (6 MA) and the MPP tracking by the ANN, trained using the PSO and
progressively grows to a maximum at noon; then, it begins to PO algorithms to determine day data. Figures 21(a) and
drop in the evening and ultimately to zero at (6 pm). 21(b) show the proposed system’s power output and tracking
With increasing solar irradiation and duration, the ac- performance using the two techniques to track the MPP for
quired power rises and decreases with decreasing power. 11 April. The maximum power obtained by the PV system
Figures 21 and 22 depict the PV system characteristics and when using the ANN trained by PSO is more remarkable
International Transactions on Electrical Energy Systems 13

Solar PV power (kw) Solar PV power (kw)

6000 6000
5000 5000
4000
3000 4000
2000
3000
1000
0 2000
−1000
1000
−2000
−3000 0
−4000
0 5 10 15 0 5 10 15

Irradiation Irradiation
PV to kW1
(a) (b)

Figure 21: Output power for (a) ANN (b) PO.

Solar PV power (kw) Solar PV power (kw)

6000
6000
5000 5000
4000
3000 4000
2000
3000
1000
0 2000
−1000
1000
−2000
−3000 0
−4000
0 2 4 6 8 10 12 14 0 2 4 6 8 10 12 14
Offset=0 Offset=0

Irradiation Irradiation
PV to kW1
(a) (b)

Figure 22: Output power for (a) ANN (b) PO.

than when using the PO approach. The ANN algorithm conditions. However, for the ANN trained by PSO, the
produced higher-quality output power than the P&O MPPT voltage value ranges between 700 Volts and more (as can be
method. The suggested ANN using the PSO algorithm re- seen in Figure 23(a)), which is supposed not to exceed 570
sponds faster to changes in irradiation, and the PO algo- and 900 Volts to achieve the inverter condition. In contrast,
rithm takes more time to respond. For example, the power the current values changed from zero to the highest value,
curve is depicted by a red line; in 5 seconds, the ANN has around 8000 amps, depending on the radiation value on that
achieved 5000 KW, which is lower than 5000 KW. The day and the temperature, as shown in Figure 23(b). On the
system’s dynamic responsiveness and steady-state power are other hand, the voltage for the PO method is slightly dif-
satisfactory, regardless of whether the irradiation is raised or ferent; as shown in Figure 24(a), it is less reliable and ef-
lowered. Compared to the PO algorithms, the ANN using fective tracking than the ANN voltage curve; however, the
the PSO algorithm has a better tracking speed, particularly voltage value ranges from 550 volts to 750 volts compared it
near the MPP. In contrast, the power output and the tracking to the ANN approach; the PO algorithm causes overshoot
performance for the proposed system using the two tech- and oscillation in the output voltage from the photovoltaic
niques to track the MPP for November 8 are depicted in system. The current curve reflects the change in climatic
Figures 22(a) and 22(b). conditions and has the current curve of the same charac-
Figures 23 and 24 show how the photovoltaic system teristics as the previous algorithm [29]. Finally, all figures
responds to two, an algorithm for changes in climatic demonstrated that the PO algorithm oscillates around the
14 International Transactions on Electrical Energy Systems

<V_PV> <I_PV>
1000 8000
900 6000
4000
800 2000
700 0
−2000
600 −4000
500 0 2 4 6 8 10 12 14
0 2 4 6 8 10 12 14 Offset=0
Offset=0
(a) (b)

Figure 23: The I-V characteristics for the ANN.

Solar PV Voltage
Solar PV Current
750 8000
700
650 6000
600 4000
550
500 2000
450 0
400
0 5 10 15 0 5 10 15
(a) (b)

Figure 24: The I-V characteristics for the PO.

Table 2: Analysis of the performance of proposed methods.

MPPT techniques Sensed variables Steady state error Tracking speed Tracking accuracy Efficiency Complexity
ANN, PSO V, I Less Faster Stable High High
P&O V, I Moderate Fast Less stable Moderate Low

MPP, resulting in power loss that can be avoided using the have proved their resistance to irradiance changing, even
ANN algorithm, which exhibits good stability, as summa- when the irradiance drops suddenly.
rized in Table 2. The ANN trained by the PSO algorithm’s performance
To sum up, it can be observed from the foregoing that was better in time response, tracking speed, and oscillation
Al Fashir station’s performance is low on the randomly than the PO algorithm and could identify the most power
selected days compared to the performance of the simu- point. Although the MPP detected values range from
lated two algorithms under the appropriate weather con- maximum powers, the tracking effectiveness of ANN al-
ditions, which is supposed to be high output. The output gorithms is excellent, indicating that the suggested approach
capacity in the two days did not exceed 4 MW during the may work well for the photovoltaic array in various climatic
specific days, while the output of the two algorithms situations. During the two days, the output capacity did not
exceeded 5 MW, especially during the afternoon. Therefore, exceed 4 MW, although the output of the two algorithms was
it is necessary to review the performance of the inverters more than 5 MW, particularly in the afternoon. Moreover,
and controllers. In contrast, the ANN trained by the PSO the MATLAB simulation demonstrated that the PO algo-
algorithm’s performance was better in time response and rithm oscillates around the MPP, resulting in power loss that
tracking speed and oscillation than the PO algorithm. It may avoid by utilizing the ANN algorithm, which exhibits
may be more effective to examine, test, and design a good stability. We recommend in advance during the test
prototype model for this task in order to accelerate the commission to check all control parts functions and seek to
growth of renewable energy research and development. improve the accuracy of the plant. The findings will raise
green power consciousness and provide a significant re-
6. Conclusion source for effectively using photovoltaic plants in the power
industry. Finally, the system was examined and constructed,
In this study, the effectiveness and performance of the Al and the performance may be investigated utilizing an ex-
Fashir photovoltaic system were examined. The simulation is perimental system in future work. Furthermore, the shadow
implemented using two cases to assess resilience of the impact is another difficulty that influences the efficiency of a
proposed system built-in MATLAB under rapidly changing photovoltaic system and will be studied in future articles.
atmospheric circumstances. The actual parameters for two This approach can be implemented for any PV system.
days were thoroughly investigated to increase the station’s
efficiency and accuracy while reducing the complexity of PV Data Availability
MPPT management. The output of the ANN trained by the
PSO and the PO algorithms are compared to the plant’s The data that support the findings of this study are available
actual performance. However, the two suggested methods from the corresponding author upon reasonable request.
International Transactions on Electrical Energy Systems 15

Conflicts of Interest [14] M. Mokhtar, M. I. Marei, and M. A. Attia, “Hybrid SCA and
adaptive controller to enhance the performance of grid-
The authors declare that they have no conflicts of interest. connected PV system,” Ain Shams Engineering Journal,
vol. 12, no. 4, pp. 3775–3781, 2021.
[15] F. Sedaghati, A. Nahavandi, M. A. Badamchizadeh, S. Ghaemi,
References and M. Abedinpour Fallah, “PV maximum power-point
tracking by using artificial neural network,” Mathematical
[1] M. Verma, H. K. Ghritlahre, and S. Bajpai, “A case study of Problems in Engineering, vol. 2012, Article ID 506709,
optimization of a solar power plant sizing and placement in 10 pages, 2012.
Madhya Pradesh, India using multi-objective genetic algo- [16] A. Gupta and S. Omveer, “Design and analysis of grid tied
rithm,” Annals of Data Science, 2021. single stage three phase P-V system,” International Journal of
[2] L. M. Halabi, S. Mekhilef, L. Olatomiwa, and J. Hazelton, Intelligent Networks, vol. 8, no. 4, pp. 3971–3978, 2021.
“Performance analysis of hybrid PV/diesel/battery system [17] F. F. Rakotomananandro, Presented in Partial Fulfillment of
using HOMER: a case study Sabah, Malaysia,” Energy Con- the Requirements for the Degree Master of Science in the
version and Management, vol. 144, pp. 322–339, 2017. Graduate School of the Ohio State University, 161 pages, The
[3] L. Alhafadhi and J. Teh, “Advances in reduction of total Ohio State University, OH, USA, 2011.
harmonic distortion in solar photovoltaic systems: a literature [18] J. Cao, “Analysis and implementation of grid-connected solar
review,” International Journal of Energy Research, vol. 44, PV with harmonic compensation,” , Florida State
no. 4, pp. 2455–2470, 2020. University73 pages, Tallahassee, Florida, 2011, , Master Thesis.
[4] H. H. Ammar, A. T. Azar, R. Shalaby, and M. I. Mahmoud, [19] M. A. Sahnoun, H. M. R. Ugalde, J.-C. Carmona, and
“Metaheuristic optimization of fractional order incremental J. Gomand, “Maximum power point tracking using P&O
conductance (FO-INC) maximum power point tracking control optimized by a neural network approach: a good
(MPPT),” Complexity, vol. 2019, Article ID 7687891, 13 pages, compromise between accuracy and complexity,” Energy
2019. Procedia, vol. 42, pp. 650–659, 2013.
[5] H. Mamur, M. A. Üstüner, and M. R. A. Bhuiyan, “Future [20] A. Bughneda, M. Salem, A. Richelli, D. Ishak, and S. Alatai,
perspective and current situation of maximum power point “Review of multilevel inverters for PV energy system Ap-
tracking methods in thermoelectric generators,” Sustainable plications,” Energies, vol. 14, no. 6, Article ID 1585, 2021.
Energy Technologies and Assessments, vol. 50, Article ID [21] W. Omran, Performance Analysis of Grid-Connected Photo-
101824, 2022. voltaic Systems, 196 pages, Phd Thesis, University of Waterloo,
[6] H. D. Tafti, A. I. Maswood, G. Konstantinou, J. Pou, and Waterloo, Canada, 2010.
F. Blaabjerg, “A general constant power generation algorithm [22] S. D. Al-Majidi, M. F. Abbod, and H. S. Al-Raweshidy, “A
for photovoltaic systems,” IEEE Transactions on Power particle swarm optimisation-trained feedforward neural
Electronics, vol. 33, no. 5, pp. 4088–4101, 2018. network for predicting the maximum power point of a
[7] Q. Fu, G. L. Cheng, F. J. Liu, and G. L. Ma, “Improvement of photovoltaic array,” Engineering Applications of Artificial
P&O MPPT method for photovoltaic system based on Intelligence, vol. 92, Article ID 103688, 2020.
adaptive prediction algorithm,” Applied Mechanics and Ma- [23] B. K. Panigrahi, A. Abraham, and S. Das, Computational
terials, vol. 263–266, pp. 2131–2137, 2012. Intelligence in Power Engineering, Springer, vol. 302, Berlin,
[8] S. Bentouba, M. Bourouis, N. Zioui, A. Pirashanthan, and Heidelberg, , 2010.
D. Velauthapillai, “Performance assessment of a 20 MW [24] S. Phommixay, M. L. Doumbia, and D. Lupien St-Pierre, “Review
photovoltaic power plant in a hot climate using real data and on the cost optimization of microgrids via particle swarm op-
simulation tools,” Energy Reports, vol. 7, pp. 7297–7314, 2021. timization,” International Journal of Energy and Environmental
[9] E. Lodhi, S. Jing, Z. Lodhi, R. N. Shafqat, and M. Ali, “Rapid Engineering, vol. 11, no. 1, pp. 73–89, 2020.
and efficient MPPT technique with competency of high ac- [25] A. Gupta and O. Singh, “Grid connected PV system with
curate power tracking for PV system,” in Proceedings of the MPPT scheme using particle swarm optimization technique,”
2017 4th International Conference on Information Science and International Journal of Intelligent Networks, vol. 02, no. 02,
Control Engineering (ICISCE), pp. 1099–1103, Changsha, July 2021.
2017. [26] S. Alepuz, S. Busquets-Monge, J. Bordonau, J. Gago,
[10] A. F. Mirza, M. Mansoor, K. Zhan, and Q. Ling, “High-ef- D. Gonzalez, and J. Balcells, “Interfacing renewable energy
ficiency swarm intelligent maximum power point tracking sources to the utility grid using a three-level inverter,” IEEE
control techniques for varying temperature and irradiance,” Transactions on Industrial Electronics, vol. 53, no. 5,
Energy, vol. 228, Article ID 120602, 2021. pp. 1504–1511, 2006.
[11] E. Lodhi, F. Y. Wang, G. Xiong et al., “A dragonfly optimi- [27] D.-L. Popa, M.-S. Nicolae, P.-M. Nicolae, and M. Popescu,
zation algorithm for extracting maximum power of grid- “Design and simulation of a 10 MW photovoltaic power plant
interfaced PV systems,” Sustainability, vol. 13, no. 19, Article using MATLAB and Simulink,” in Proceedings of the 2016 IEEE
ID 10778, 2021. International Power Electronics and Motion Control Conference
[12] M. S. Ngan and C. W. Tan, “Multiple peaks tracking algorithm (PEMC), pp. 378–383, Varna, Bulgaria, September 2016.
[28] X. Lue, “Harmonic analysis of photovoltaic generation in
using particle swarm optimization incorporated with artificial
distribution network and design of adaptive filter,” Interna-
neural network,” International Journal of Electronics and
tional Journal of Computing and Digital Systems, vol. 9, no. 1,
Communication Engineering, vol. 5, no. 10, 7 pages, 2011.
pp. 77–85, 2020.
[13] N. Khaldi, H. Mahmoudi, M. Zazi, and Y. Barradi, “The
[29] S. Messalti, A. Harrag, and A. Loukriz, “A new variable step
MPPT control of PV system by using neural networks based
size neural networks MPPT controller: review, simulation and
on Newton Raphson method,” in Proceedings of the 2014
hardware implementation,” Renewable and Sustainable En-
International Renewable and Sustainable Energy Conference
ergy Reviews, vol. 68, pp. 221–233, 2017.
(IRSEC), pp. 19–24, Ouarzazate, Morocco, October 2014.