Received January 28, 2020, accepted February 14, 2020, date of publication February 21, 2020, date of current

version March 3, 2020.

Digital Object Identifier 10.1109/ACCESS.2020.2975742

A Hybrid Evolutionary-Based MPPT for

Photovoltaic Systems Under Partial
Shading Conditions
MANSI JOISHER1 , DHARAMPAL SINGH2 , SHAMSODIN TAHERI 3 , (Senior Member, IEEE),
DIEGO R. ESPINOZA-TREJO 4 , EDRIS POURESMAEIL 5 , (Senior Member, IEEE),
AND HAMED TAHERI 6
1 Department of Electrical and Electronic Engineering, National Institute of Technology, Surathkal, Mangalore 575025, India
2 Department of Electrical Engineering, Punjab Engineering College, Chandigarh 160012, India
3 Department of Computer Science and Engineering, Université du Québec en Outaouais, Gatineau, QC J8X 3X7, Canada
4 Department of Renewable Energy, Universidad Autónoma de San Luis Potosí, San Luis Potosí 78700, Mexico
5 Department of Electrical Engineering and Automation, Aalto University, 00076 Espoo, Finland
6 Current, powered by GE, Montreal, QC H8T 3M6, Canada

Corresponding author: Shamsodin Taheri ([email protected])

This work was supported by the Natural Sciences and Engineering Research Council of Canada (NSERC), Discovery Grants Program,
RGPIN-2015-05175.

ABSTRACT Under partial shading conditions (PSCs), photovoltaic (PV) system characteristics vary and
may have multiple power peaks. Conventional maximum power point tracking (MPPT) methods are unable
to track the global peak. In addition, it takes a considerable time to reach the maximum power point (MPP).
To address these issues, this paper proposes an improved hybrid MPPT method using the conventional
evolutional algorithms, i.e., Particle Swarm Optimization (PSO) and Differential Evaluation (DE). The main
feature of the proposed hybrid MPPT method is the advantage of one method compensates for shortcomings
of the other method. Furthermore, the algorithm is simple and rapid. It can be easily implemented on a
low-cost microcontroller. To evaluate the performance of the proposed method, MATLAB simulations are
carried out under different PSCc. Experimental verifications are conducted using a boost converter setup,
an ET-M53695 panel and a TMS320F28335 DSP. Finally, the simulation and hardware results are compared
to those from the PSO and DE methods. The superiority of the hybrid method over PSO and DE methods is
highlighted through the results.

INDEX TERMS Photovoltaic systems, maximum power point tracking, partial shading condition, particle
swarm optimization, differential evaluation.

I. INTRODUCTION Due to these barriers, PV systems generate lower power [1].

In recent years, photovoltaic (PV) systems integrated into More importantly, PV systems show a lower performance
power grids have been gaining popularity as one of the most when the solar irradiance is not uniformly distributed over the
promising and reliable energies among existing renewable PV array surface, known as partial shading phenomenon [2].
energy sources. PV sources offers several advantages in terms The latter changes the power-voltage (P-V) characteristics
of being clean, renewable and low maintenance. However, of PV systems by generating multiple local power maxima.
the low efficiency of PV systems often due to their nonlinear It remains a challenge of global optimization to ensure that
electrical characteristic and variable atmospheric conditions PV systems operate at their global maximum power point
remains a great challenge. Therefore, in order to overcome (GMPP) rather than at the local ones [3], [4]. Over recent
this main drawback a PV system needs to operate at its years, several maximum power point tracking (MPPT) tech-
maximum power point (MPP). The efficiency of PV systems niques in combination with power electronic devices have
is considerably affected by local atmospheric conditions such been proposed to deliver maximum power from the PV
as moving clouds, dust, neighbouring buildings and trees. array. These techniques vary in some general parameters
such as complexity, accuracy, cost and speed. Among these,
The associate editor coordinating the review of this manuscript and Hill climbing (HC) [5] and perturb and observe (P&O) [6]
approving it for publication was Tariq Masood . are the most common used algorithms due to the ease of

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see http://creativecommons.org/licenses/by/4.0/
VOLUME 8, 2020 38481
 M. Joisher et al.: Hybrid Evolutionary-Based MPPT for PV Systems PSCs

implementation and their simple control mechanisms. These rapid convergence and it needs few control parameters [21],
two algorithms showed a similar fundamental approach to [22]. However, in DE particles compete for survival while
achieve the MPP. The HC algorithm operates by periodically the winners hardly retain enough history. Moreover, there is
perturbing the duty cycle for a power converter, and the P&O no cooperation among particles to find the GP consequently,
algorithm works with a perturbation in the operating voltage an increase in the time computation is probable. PSO is a
of the PV system. The output power is a determining factor in population-based search method, which searches the MPP
increasing or decreasing control parameters (voltage or duty through a swarm of particles where each particle is considered
cycle) to reach the MPP. Despite of simplicity in operation of as a candidate solution. The particles in PSO cooperate to
theses algorithms, the steady state system oscillates around find the global best position [23]. The PSO method requires
the MPP. The selection of a small step size of the pertur- no gradient information of function to be optimized. It has
bation can improve the oscillations while it decreases the a simple, effective and fast metaheuristic approach to find
speed of the system. Some improved P&O techniques have the GP under PSCs. However, several improvements have
been proposed to overcome these drawbacks through variable been made to the conventional PSO technique to enhance its
perturbation step size [7], [8]. However, the major drawback efficiency. In [24] similar to HC, the improved PSO works
of these algorithms is that they are unable to find the global with direct duty cycle with a faster tracking speed and low
peak (GP) under partial shading conditions (PSCs). oscillations at the MPP under PSCs. The method presented
The Incremental Conductance (IC) is another commonly in [25] is a system-independent MPPT algorithm which sorts
used MPPT technique [9], [10]. The MPP is reached when the obtained particle positions to avoid a large voltage stress
the slope of the PV power curve is zero. Similar to the on the power switch due to the sudden change of duty cycle.
P&O method, the IC suffers from a trade-off between the The authors in [26] proposed a deterministic PSO-based
accuracy and speed of the system to reach the MPP, as well MPPT under PSCs. The control structure of this MPPT can
as incapability in tracking the GP under PSCs. be simplified since the random number in the acceleration
Fractional open-circuit voltage (FOCV) [11] and Frac- factor of the conventional PSO is removed. However, two
tional short-circuit current (FSCC) [12] algorithms have been separated flowcharts are implemented to determine the local
found simple methods to track the MPP of a PV system. The and global modes compared to the conventional PSO with a
FOCV is based on a linear relationship between open-circuit unique algorithm. A hybrid method, which combines P&O
voltage and the MPP voltage. In the same way, the FSCC is and PSO is proposed in to track the MPP in two stages [27].
presented as a ratio of short-circuit current to the MPP cur- In the first stage, the P&O method is employed to search for
rent. Due to the approximate aforementioned relationships, the first local maximum power point (LMPP) and then in the
the PV system cannot find a true MPP to operate. In spite second stage, the PSO is used to search the GMPP. The time
of the above-mentioned techniques, several methods such as chosen to apply the associated algorithm is a determining
Ripple Correlation Control (RCC) [13], the slide mode con- factor, in particular, when the partial shading occurs. Despite
trol [14], [15] and dP/dV feedback control [16] have been also the remarkable accuracy of the above-mentioned PSO meth-
proposed to achieve the MPP of the PV array. These methods ods compared to the conventional one under PSCs, the time
often fail to track the GP under nonuniform insolation as the required for convergence is still long. In PSO each particle is
electrical characteristics of PV exhibit multiple peaks. influenced by its personal best performance and that achieved
In order to improve the above-mentioned drawbacks, sev- by the best particle in its neighborhood throughout its life-
eral methods in the family of soft computing (SC) techniques time. However, the particles in PSO are not eliminated when
such as artificial intelligence approach and evolutional algo- they meet a new worth position. Thus, longer computation
rithms have been attracting considerable interests recently. time for large search space, particularly under various par-
In the field of artificial intelligence techniques, fuzzy logic tial shading is inevitable. Another issue regarding the PSO
controllers (FLC) [17], [18], artificial neural network (NN) technique is that it can be trapped at the local peak in a
[19], [20] methods have shown effective solution in dealing high-dimensional space and has a low convergence rate [28]–
with the nonlinear characteristics of the current-voltage (I-V) [30]. To overcome these main drawbacks, this paper proposes
curve, in particular, under PSCs. However, they require exten- an improved MPPT method based on a synergism of PSO
sive computation. and DE, called the PSO-DV algorithm. In the PSO-DV, the
The evolutional algorithm (EA) technique as a stochastic mutation and crossover processes in DE method are used to
optimization method appears to be very efficient in tracking generate a trial vector. Then, using merit criterion, dislike the
the MPP. The EA is able to find the GP regardless of inso- PSO, the particle is moved to the new position only if this
lation pattern as it is based on search optimization. Among location meets a better fitness value. Hence, the computation
various EA techniques, differential evolution (DE) and par- time could be greatly reduced, in particular, under PSCs.
ticle swarm optimization (PSO) has been gaining popularity Moreover, to avoid the particles from being trapped at one of
in tracking the MPP under PSCs. Several MPPT algorithms the local peaks, a condition is used in the proposed algorithm.
based on PSO and DE techniques have been presented to date The particle escapes from the LMPP by a random mutation
to recognize the GP among multiple local peaks. The DE and keeps the swarm moving. In fact, the proposed PSO-DV
offers several advantages as it is able to find the GP, it has MPPT i) benefits from features of PSO and DE for searching

38482 VOLUME 8, 2020

M. Joisher et al.: Hybrid Evolutionary-Based MPPT for PV Systems PSCs

the GP under PSCs, ii) decreases greatly the computation Step 1: Initialization
time, and iii) copes with dynamic change in irradiance levels The optimization process of DE starts through an ini-
under large fluctuations of insolation. tial
h population of D-dimensional i parameter vectors xG i =
The rest of the paper is organized as follows. An overview x1,i ,x2,i , . . .xj,i , . . .xD,i . Individuals x1j,i are initialized
G G G G
of PSO and DE algorithms is presented in Section II. The syn- randomly in the certain range with a lower and upper limit,
ergy of PSO and DE is developed in Section III. The theory xL andxH as follows:
of the PSO-DV is implemented in Section IV. The simula-
tion verification is presented in Section V. The experimental x1j,i = xL + rand[0, 1](xH − xL ) (3)
results are illustrated in Section VI. Finally, the concluding
= xL,1 ,xL,2 , . . .xL,D and xH = [xH,1 ,xH,2 , . . .
 
remarks are presented in Section VII. where xL
xH,D ].
II. OVERVIEW OF PSO AND DE ALGORITHMS Step 2: Mutation
PSO and DE algorithms have been known as popular tech- In order to generate a donor vector vG i , mutation process
niques of EA. Recently, these techniques have been gaining should be applied to each parameter vector xG i in the way
much attention due to their ability in optimizing real-valued that, three vectors (xG , x G , xG ) are randomly selected in the
r1 r2 r3
nonlinear and multi-modal objective functions. As these tech- range [1, NP] where NP is the population size. It should
niques are based on search optimization, the GP could be be mentioned that the indices i, r1 , r2 and r3 are distinct.
tracked with a reasonable convergence time and a better The mutated vector vG i is generated by adding the weighted
dynamic response than the conventional methods. difference between two vectors, randomly chosen, to a third
vector as follows:
A. PSO ALGORITHM
vG G G G
i = xr1 + F(xr2 − xr3 ) (4)
PSO is a stochastic optimization method, inspired by the
behavior of a flock of birds or a school of fish, developed where F is a mutation scaling factor that scales the difference
in 1995 [31]. In fact, the PSO as a metaheuristic approach of two vectors, which is usually selected between 0 and 1.
is used to optimize a function that is difficult to express Step 3: Crossover
analytically. In PSO, the global position is searched by a num- The crossover operation involves generating a trail vector,
uGi = [u1,i ,u2,i , . . .uj,i , . . .uD,i ] through mixing the target
ber of agents (particles) with a continually updated velocity. G G G G
The movement of each particle around in the search space vector xG G
i with the mutated vector vi . The DE uses commonly
is controlled by its own best position and the globally best two kinds of crossover schemes, i.e., binomial and exponen-
position found by the entire particles so far. The position and tial [32]. The binomial crossover method, used in this work,
velocity of the ith particle in the swarm are determined as yields the following condition to produce the trail vector.
follows: (
vGj,i if rand(0, 1) < CR
xk+1 = xki + vk+1 (1) uG
j,i = (5)
i i xGj,i otherwise
vk+1
i = ωvki + c1 r1 (pbesti − xki )+c2 r2 (gbest − xki ) (2)
where CR is known as crossover constant, similar to F, is a
where xki is the position of individual i at iteration k, vki is control parameter of DE.
the velocity of individual i at iteration k, ω is the inertia Step 4: Evaluation and selection
weight, c1 and c2 are the acceleration coefficients, r1 and r2 An important point that should be taken into account in
are the random positive numbers between 0 and 1, pbesti is the DE method is to keep the population size constant during
personal best position of individual i and gbest represents the the operation. The DE method involves competing between
global best position found so far in the community. the target and trial vectors to take the place for the next
generation. In fact, the trial vector will replace the target
B. PSO ALGORITHM vector (parent vector) if it achieves the best fitness value,
DE, which is a member of the genetic algorithm (GE), was otherwise the target vector remains in the population. Hence,
introduced by Storn and Price in 1995 [32]. The DE is known it ensures that the population either gets better fitness value or
as a stochastic, population-based optimization algorithm. The continues with the same target vector, but never deteriorates.
DE technique starts with the initialization of a random pop- The selection stage can be expressed by the following condi-
ulation and then the mutation operation is applied on the tion that the objective function f(x) requires to be minimized.
individuals of the current population, called target vectors, (
uG i ) < f(xi )
if f(uG G
to produce a mutated vector. Then, the crossover operation xiG+1
= G i (6)
generates a new vector (trial vector). The trial vector can xi otherwise
replace the target vector in the next irritation if the trial vector
acquires a better fitness value than the target vector. The III. SYNERGISM OF PSO AND DE
optimization process of DE could be presented briefly as As mentioned, the main feature of the PSO is the cooperation
follows: among the population individuals. In fact, each individual

VOLUME 8, 2020 38483

 M. Joisher et al.: Hybrid Evolutionary-Based MPPT for PV Systems PSCs

FIGURE 2. Movement of particles in PSO-DV.

FIGURE 1. Movement of particles in PSO.

According to the obtained velocity vector, vk+1

i a new trial
determines its destination based on its personal best expe- position Ti is generated as follows:
rience and that globally achieved in the population. In PSO
algorithm, the values of r1 , r2 and ω can affect the direc- Ti = xki + vk+1
i (9)
tion of a particle to the MPP as they conduct a particle
toward the direction of pbesti , gbest and its previous direction. To select the next position of each particle the following
Since the values of r1 and r2 are random, distributed within condition with the objective function f (x) is proposed.
(
[0,1], the next position of the particle might not be better Ti if f(Ti ) < f(xki )
k+1
than the pbesti (Fig. 1). Moreover, in PSO the particles are xi = k (10)
xi otherwise
not eliminated even when they experience the worst fitness.
Thus, since the particles remain in the memory of PSO, Therefore, for each irritation, the new velocity is deter-
it wastes the limited computational resources, consequently mined and the particle either reaches a better position in the
a slower speed in convergence. On the other hand, the DE search space or keeps its previous location. Thus, the current
algorithm employ selection and mutation operators to forces position of the particles is the best ones that they have ever
the individual agent to find a better fitness value to survive experienced. Hence, the term (pbesti − xki ) in PSO is elimi-
in the next generation. The main characteristic of DE is to nated. Fig. 2 shows the typical movement of a particle in the
keep the competition in the population while the winning PSO-DV. As observed from Fig. 2, the vector pbesti − xki is
particles hardly keep sufficient history. Therefore, the advan- eliminated as the actual position of the particle is its personal
tage of one method can compensate for the shortcomings of best fitness value.
the other technique. Hence, a new algorithm was developed Sometimes for a number of iterations, a particle may get
based on a synergy of PSO and DE, called PSO-DV [33]. stagnant at any point the search space (locals) as follows:
In order to circumvent the above-mentioned weakness of
each method, a differential operator, borrowed from DE, x1+K
i = x2+K
i = x3+K
i . . . = xK+N
i (11)
in the mutation stage is coupled with the velocity update To avoid this problem, the particle is shifted by a random
scheme in PSO. The operator is invoked on the position mutation to a new location through the following equation.
vectors of two randomly chosen individuals, different from
their best fitness. xk+N+1
i = xmin + rand[0, 1](xmax − xmin ) (12)
Thus, in the new algorithm, individuals are prohibited from
where xmax and xmin are the permissible bounds of the search
circulating in the useless regions of the search space. Similar
area and N is the maximum iteration number.
to what we have in the DE, for a given particle i, two random
The new algorithm offers both respective cooperative and
distinct particles j and k are selected. Hence, a difference
competitive features of PSO and DE. In the PSO-DV the
vector y can be defined as follows:
promising individuals share their experiences, which leads to
−
→
y =−
→
xk − −
→
xj (7) a faster convergence. Moreover, worse individuals far away
from the global optima are prohibited to participate in the
Then, the velocity vector in (2) is modified and then pre- next irritation. In this way, the search area becomes more
sented in the selection condition as follows: concentrated, close to the global optima leading in saving
( the computational time. Furthermore, the mutation operator
ωuk +∝ y+c2 r2 gbest − xki if rand(0, 1) < CR


k+1
vi = k i incorporates to PSO to increase diversity of the population
vi otherwise that enables the particles to escape from local trap.
(8)
IV. IMPLEMENTATION OF PSO-DV TO MPPT
where CR is a random value in (0,1). The proposed PSO-DV described in the previous section is
Thus, it can be seen that the velocity vector created by the applied to the MPPT algorithm to track the MPP, in par-
personal best experience is replaced with differential vector ticular, under PSCs. To evaluate the effectiveness of pro-
borrowed from the DE method. posed PSO-DV algorithm, a simple block diagram (Fig. 3)

38484 VOLUME 8, 2020

M. Joisher et al.: Hybrid Evolutionary-Based MPPT for PV Systems PSCs

FIGURE 3. MPPT connection to DC-DC convertor.

consisting of a PV panel, a DC-DC converter, a load and a

controller in which the proposed MPPT technique is imple-
mented is presented. The MPPT controls the converter oper-
ation through the duty cycle varied by measured voltage and
current at the final output. This circuit simplification reduces
the cost and complexity of the system rather than where
individual MPPT along with associated voltage and current
sensors, located at the output of each panel. In order to imple-
ment the PSO-DV method into the MPPT algorithm, the duty
cycles di with a population consisting of Np individuals are
defined.

xki = dki = [d1 , d2 , d3 , . . . ,dNP ] (13)

Moreover, the following objective function is defined.

 
P dki > P(dk−1
i ) (14)

The PSO-DV algorithm is based on the PSO that the

velocity vector is perturbed by the mutation process, bor-
FIGURE 4. Flowchart of the PSO-DV technique.
rowed from DE. Fig. 4 shows the flowchart of the proposed
PSO-DV MPPT that includes the following general steps:
Step 1: Parameters values selection Step 5: Velocity calculation of particles by differential
The values of required parameters in the PSO-DV includ- operator
ing population size, maximum irritation, learning factor and When the fitness values of particles are determined,
inertia weight are selected. the updated velocity of particles should be determined
Step 2: Particles initialization by three terms, generated by the velocity of the particle,
The particles can be initialized on a fixed position in the the mutation process (borrowed from the DE algorithm),
range of dmin to dmax that dmin and dmin are the respective and the information obtained from the global best position.
maximum and minimum duty cycles. Empirical observation Equations (8-10) are represented in terms of duty cycle as the
has shown that the MPP on the P-V curve occurs at 80% of position of each particle as follows:
the open voltage of the PV module. (
Step 3: Fitness evaluation ωuki +∝ y + c2 r2 (gbest − dki ) if rand (0, 1)< CR
vk+1 =
The particle i, which is the output of digital controller, i vki otherwise
controls the output voltage and current through the PWM (15)
command. The fitness value PPV of each particle (duty cycle) Ti = dki + vk+1 (16)
i
then can be determined through the measured voltage and (
current. T i if f(Ti ) < f(dki )
dk+1
i = (17)
Step 4: Update personal and global best positions dki otherwise
According to (10), the current position of the particle in
PSO-DV method is the best position it has ever experienced. Step 6: Escape from local peaks
Thus, in this step, the best fitness value of the particles in the If the particle (duty cycle) oscillates around a point for a
community for each irritation is selected. number of iterations, the particle can escape from the LMPP

VOLUME 8, 2020 38485

 M. Joisher et al.: Hybrid Evolutionary-Based MPPT for PV Systems PSCs

TABLE 2. Parameters of the PV system.

FIGURE 5. Simulation schematic model.

TABLE 1. Electrical parameters of the ET-M53695 panel at STC.

by a random mutation as follows:

d1+K
i = d2+K
i = d3+K
i . . . = dK+N
i (18)
dk+1
i = dmin + rand[0, 1](dmax − dmin ) (19)
FIGURE 6. Characteristics curves for Case 1.
where N is the maximum number of iterations.
Step 7: Convergence detection
The PSO-DV will stop if the maximum number of irrita- In order to justify the proposed PSO-DV method, three
tions is reached. Thus, the obtained position is gbest. partial shading cases are examined. The results are compared
Step 8: Partial shading verification with the conventional PSO and DE methods. Fig. 6 represents
The operating point (gbest) obtained by the MPPT changes the corresponding I-V and P-V curves for Case 1. Out of
with environmental conditions as well as the load connected the 3 PV panels under investigation, 2 panels were shaded in
across the panel. In such conditions, the new or initialized this case. This results in 2 peak values in which the leftmost
particles should start searching the MPP. Thus, the algorithm peak is the GP. The maximum power is 22.17W at 17.26 V.
is reinitialized whenever the following condition is satisfied. Fig. 7 illustrates the simulation results for the PSO, DE and
proposed PSO-DV methods when the uniform irradiance con-
Ppv,new − Ppv,last
≥ 1P(%) (20) dition changes to Case 1 at t = 0.5 s.
Ppv,last It can be observed that the PSO algorithm requires around
where Ppv,last is the power at the MPP of the last operating 5 seconds to stabilize at the GP, on the other hand, the DE
point. algorithm is not stabilized till 9 seconds, whereas, the pro-
The proposed PSO-DV reduces the computational time posed algorithm PSO-DV stabilizes at the GP in 1 second.
rather than the conventional PSO as the instant value of each Given that there is only a single local peak for the output
particle (duty cycle) is the best one that it has had so far. curve, with a considerable difference between GMPP and
Thus, unlike the conventional PSO, the PSO-VD prohibits the LMPP, it is easier to track this GMPP through PSO method
particles from visiting the unnecessary positions by the dif- with a proper initialization of particles. By contrast, the ini-
ferential operator borrowed from the DE. In addition, in order tialization is completely random in DE and hence can be used
to avoid from oscillating around a local optimum, an online to evaluate all ranges of search space but the search quality
verification condition is added to the conventional PSO. The is degraded in this case. For this case, results prove that the
particle can be shifted to a new location by a random mutation proposed algorithm is able to track the GMPP faster than the
when it gets stagnant at local optima. DE and PSO algorithms. The proposed algorithm experiences
less power oscillations.
V. SIMULATION VERIFICATION Fig. 8 represents the characteristic curves for Case 2. In this
To verify the correctness of the proposed method, simu- particular case, all the 3 PV panels are partially shaded to
lation analysis is conducted using MATLAB/SIMULINK. a certain degree. This results in 3 peaks in the P-V curve,
Fig. 5 illustrates the schematic of the PV system used in the maximum power, in this case, is 135W at 37V. The local
this study. This consists of a DC-DC boost converter, a PV peaks are located at 126W and 88W respectively.
array, a battery pack and the MPPT controller. The converter This condition differs from the previous condition as
is designed for continuous inductor current mode. The key the difference between GMPP and one of the LMPP is
specifications of the converter and PV module are shown lesser, which this makes it difficult for the conventional
in Table 1 and Table 2. techniques to track the GMPP. In the PSO algorithm as

38486 VOLUME 8, 2020

M. Joisher et al.: Hybrid Evolutionary-Based MPPT for PV Systems PSCs

FIGURE 9. Simulation results for case 2. (a) PSO. (b) DE. (c) PSO-DV.

FIGURE 7. Simulation results for case 1. (a) PSO. (b) DE. (c) PSO-DV.

FIGURE 8. Characteristics curves for Case 2.

FIGURE 10. Characteristics curve for case 3.

the iterations increase, the search space is narrowed down.
Hence, the particles get trapped at the local peaks as already
stated in [28]–[30]. It can be seen from the PSO simulation peak, 92 W. The middle peak, a local peak is at 80 W.
result, the particles are stuck at the local peak at 125 W. The difference between the middle local peak and the GP
However, the conventional PSO algorithm can be used in this is less and hence the conventional techniques get trapped
case by initializing the particle near the MPP. This makes the at this local peak. It is verified from the simulation results
algorithm-specific to a particular condition and hence, will that particles in PSO are not able to escape local peaks. PSO
fail to track GMPP in other PSCs. algorithm can be used to track GMPP in this case by setting
On the other hand, DE algorithm is able to track the GMPP some limitations on the searching area or moving direction,
in this case but the results prove the proposed algorithm which reduces the searching time as well as high-tracking
PSO-DV reached the GMPP faster. However, the proposed efficiency. However, the PSO method loses its randomness
algorithm PSO-DV uses a random mutation loop to escape and then loses its inherent advantages. Similar to the previous
local peaks. Hence, it experiences power oscillation. The case the DE algorithm is able to track the GMPP but the
proposed method is able to achieve the MPP faster compared performance is very poor. The proposed algorithm is able
to other methods. to escape local peaks due to random mutation. However,
Case 3 is demonstrated in Fig. 10, there are 3 peak val- the result of the proposed algorithm shows that it evaluates all
ues for the output curve, the GP is located at the rightmost ranges of search space faster than PSO and DE algorithms.

VOLUME 8, 2020 38487

 M. Joisher et al.: Hybrid Evolutionary-Based MPPT for PV Systems PSCs

FIGURE 12. Experimental setup of PV systems and converter.

TABLE 4. Parameters of the PV system.

FIGURE 11. Simulation results for case 3. (a) PSO. (b) DE. (c) PSO-DV.

TABLE 3. Electrical parameters of the ET-M53695 panel at STC.

frequency. The timer interrupt frequency determines how
often the duty cycle of the PWM signal is updated. The timer
interrupt frequency is kept at 10 Hz or every 100 ms. The
selection of this frequency depends on the time constant of
the specific PV system. In fact, the PV system should reach
a steady-state operation before another MPPT cycle begins.
The timer interrupt frequency is critical as it determines the
VI. EXPERIMENTAL VALIDATION update frequency of the MPPT. It must also ensure that the
The experimental tests were conducted using a PV array code does not overrun and cause unpredictable behaviour.
consisting of two series-connected ET-M53695 PV mod- Natural PSCs often occurred as a result of the shadow of
ules as shown in Fig. 12. The electrical parameters of a nearby building. Artificial PSCs were also created using
ET-M53695 Panel at STC are listed in Table 3. The terminal plastic sheets with different transparencies. The PV charac-
voltage of the PV array is controlled by duty cycle of the teristics were monitored using HT Instruments I-V 400W
IGBT switch of the boost converter to achieve the MPP. The PV Panel Analyzer and irradiance meter test kit. Similar
PV array is connected to a load via the boost converter. The to the simulation section, three algorithms of PSO, DE and
parameters of the experimental setup are listed in Table 4. the proposed DV-PSO are experimentally tested for different
A voltage sensor and a current sensor are used to measure scenarios of PSCs and the power graphs approaching the MPP
the PV array voltage and current. The algorithms are imple- are captured on code composer studio as follows. For Case 1,
mented using the Texas Instruments TMS320F28335 DSP. as shown in Fig.13, there are 2 peaks in the P-V curve and the
The platform for software development is the Code Composer leftmost peak is the global maxima, at 11 W whereas the local
Studio (CCS 8.1.0). The control programs were developed in peak is at 5.6 W. Fig 14. illustrates the experimental results
C environment. for Case1. It is observed that the PSO and DE algorithms
The enhanced ePWM1 module of DSP is used as the are stuck at the local maxima of 5W, whereas the proposed
internal clock base for the sampling frequency. The ePWM1A algorithm PSO- DV is able to reach the global maxima at
unit is used to trigger the ADC conversion sequence and 11 W, 7.7V.
ePWM1B is used to trigger the IGBT switch in the boost The proposed algorithm is able to escape all the local peaks
converter circuit. The ADC sampling is triggered by the due to the random mutation condition. Whenever the number
PWM on every first event with a sampling rate of 20 kHz. of iterations is equal to the maximum number of iterations
The CpuTimer0 of the DSP is used to set the MPPT controller and the particle is oscillating around a particle position,

38488 VOLUME 8, 2020

M. Joisher et al.: Hybrid Evolutionary-Based MPPT for PV Systems PSCs

FIGURE 13. Characteristics graphs for case 1. FIGURE 15. Characteristics graphs for case 2.

FIGURE 16. Experimental results for case 2. (a) PSO. (b) DE. (c) PSO-DV.

FIGURE 14. Experimental results for case 1. (a) PSO. (b) DE. (c) PSO-DV.

stuck at an untrue point at 17 W. The proposed algorithm DV-

the random mutation loop is initialized. Hence, in this case, PSO is able to track the GMPP as fast as the PSO algorithm,
the proposed algorithm faces more power oscillations before and unlike PSO it is not a case-specific algorithm. However,
stabilizing at a global peak. after every 40 samples, the algorithm experiences random
Case 2 is shown in Fig. 15, there are 3 peaks in the P-V mutation to ensure particles are not stuck at a local peak.
curve of the system. The middle peak is the global maxima To ensure a stable graph, once the particles have reached a
at 20W. The difference between the voltage at global maxima global peak, the maximum number of iterations can be set
and local maxima is considerable and hence this is an easy to 1000 and the random mutation loop can be accordingly
situation for both PSO and DE techniques. It is observed modified.
from the hardware verification results shown in Fig.16. PSO Case 3 demonstrated in Fig. 17 is different from the previ-
algorithm is easily able to reach the GP as the initialization of ous case as the difference between the power at the GP and
particles was set in this particular way. DE is able to escape the local peak is less. This makes it difficult for conventional
the local peaks and reach global maxima initially but later is techniques to track the GP. On careful observation, it is noted

VOLUME 8, 2020 38489

 M. Joisher et al.: Hybrid Evolutionary-Based MPPT for PV Systems PSCs

changing the maximum number of iterations. As explained

earlier this is the reason, PSO-DV algorithm is able to escape
all the local peaks in every situation.

VII. CONCLUSION
This paper is a continuation of usage of computational algo-
rithms to find the GP of PV systems under PSCs. A high-
FIGURE 17. Characteristics graphs for case 3. performance MPPT hybrid algorithm based on PSO and DE
algorithms is proposed. Through a sequential mathematical
overview of PSO and DE algorithms, a hybrid algorithm
PSO-DV is proposed. The main features of PSO and DE
are combined to overcome their drawbacks. The proposed
hybrid MPPT uses a random mutation loop to escape from
all the local peaks. The triggering of this loop depends upon
the maximum number of iterations. The performance of the
hybrid MPPT is verified using both simulation and hardware
setup. The proposed methodology is tested for 3 different
PSCs. The proposed algorithm is an advancement in this field
that makes the system more robust and improves the overall
computational speed. Contrary to the conventional PSO and
DE techniques, it can find the true GMPP. This algorithm
successfully copes with dynamic change in irradiance levels
and is applicable for real-time conditions.

REFERENCES
[1] M. Z. Shams El-Dein, M. Kazerani, and M. M. A. Salama, ‘‘Optimal pho-
tovoltaic array reconfiguration to reduce partial shading losses,’’ IEEE
Trans. Sustain. Energy, vol. 4, no. 1, pp. 145–153, Jan. 2013.
[2] H. Patel and V. Agarwal, ‘‘MATLAB-based modeling to study the effects
of partial shading on PV array characteristics,’’ IEEE Trans. Energy Con-
vers., vol. 23, no. 1, pp. 302–310, Mar. 2008.
[3] S. Hosseini, S. Taheri, M. Farzaneh, and H. Taheri, ‘‘A high-performance
shade-tolerant MPPT based on current-mode control,’’ IEEE Trans. Power
Electron., vol. 34, no. 10, pp. 10327–10340, Oct. 2019.
[4] E. Karatepe, Syafaruddin, and T. Hiyama, ‘‘Simple and high-efficiency
photovoltaic system under non-uniform operating conditions,’’ IET Renew.
Power Gener., vol. 4, no. 4, pp. 354–368, 2010.
[5] W. Zhu, L. Shang, P. Li, and H. Guo, ‘‘Modified hill climbing MPPT
algorithm with reduced steady-state oscillation and improved tracking
efficiency,’’ The J. Eng., vol. 2018, no. 17, pp. 1878–1883, Nov. 2018.
[6] A. K. Abdelsalam, A. M. Massoud, S. Ahmed, and P. N. Enjeti,
‘‘High-performance adaptive perturb and observe MPPT technique
for photovoltaic-based microgrids,’’ IEEE Trans. Power Electron., vol. 26,
no. 4, pp. 1010–1021, Apr. 2011.
FIGURE 18. Experimental results for case 3. (a) PSO. (b) DE. (c) PSO-DV.
[7] S. Kumar Kollimalla and M. Kumar Mishra, ‘‘Variable perturbation size
adaptive P&O MPPT algorithm for sudden changes in irradiance,’’ IEEE
Trans. Sustain. Energy, vol. 5, no. 3, pp. 718–728, Jul. 2014.
[8] G. Petrone, G. Spagnuolo, and M. Vitelli, ‘‘A multivariable perturb-and-
that the right-hand side peak is the global maxima at around Observe maximum power point tracking technique applied to a single-
14 W, 30 V. The leftmost peak is a local peak at around stage photovoltaic inverter,’’ IEEE Trans. Ind. Electron., vol. 58, no. 1,
13.5 W, 16 V. pp. 76–84, Jan. 2011.
[9] M. A. Elgendy, B. Zahawi, and D. J. Atkinson, ‘‘Assessment of the incre-
Fig. 18 represent the hardware verification results for mental conductance maximum power point tracking algorithm,’’ IEEE
Case 3. It is observed from the results that power for all Trans. Sustain. Energy, vol. 4, no. 1, pp. 108–117, Jan. 2013.
the algorithm is almost the same around 13.5W. However, [10] K. Soon Tey and S. Mekhilef, ‘‘Modified incremental conductance algo-
rithm for photovoltaic system under partial shading conditions and load
the voltage is in the case of PSO and DE algorithms are variation,’’ IEEE Trans. Ind. Electron., vol. 61, no. 10, pp. 5384–5392,
16.05 V and 17.7 V respectively. The voltage of the PV Oct. 2014.
system in PSO-DV algorithm is 26.8 V. This proves that the [11] O. Lopez-Lapena and M. Penella, ‘‘Low-power FOCV MPPT controller
with automatic adjustment of the sample&hold,’’ Electron. Lett., vol. 48,
first 2 algorithms are stuck at the left-hand side peak which is no. 20, pp. 1301–1303, 2012.
a local peak. The proposed algorithm is able to approach the [12] H. Ahmed Sher, A. Faisal Murtaza, A. Noman, K. E. Addoweesh,
global maxima faster than conventional methods. The algo- K. Al-Haddad, and M. Chiaberge, ‘‘A new sensorless hybrid MPPT
algorithm based on fractional short-circuit current measurement and
rithm experience power oscillation after every 40 samples P&O MPPT,’’ IEEE Trans. Sustain. Energy, vol. 6, no. 4, pp. 1426–1434,
due to the random mutation loop. This can be adjusted by Oct. 2015.

38490 VOLUME 8, 2020

M. Joisher et al.: Hybrid Evolutionary-Based MPPT for PV Systems PSCs

[13] A. Costabeber, M. Carraro, and M. Zigliotto, ‘‘Convergence analysis and MANSI JOISHER is currently pursuing the
tuning of a sliding-mode ripple-correlation MPPT,’’ IEEE Trans. Energy bachelor’s degree in electrical engineering with
Convers., vol. 30, no. 2, pp. 696–706, Jun. 2015. the National Institute of Technology, Mangalore,
[14] E. Mamarelis, G. Petrone, and G. Spagnuolo, ‘‘Design of a sliding-mode- India. Her main research interests include renew-
controlled SEPIC for PV MPPT applications,’’ IEEE Trans. Ind. Electron., able energy and power electronics. She received
vol. 61, no. 7, pp. 3387–3398, Jul. 2014. the Mitacs Globalink Research Award to attend
[15] C. Cabal, F. Guinjoan, C. Alonso, L. Martínez-Salamero, and L. Séguier, the Université du Quebec en Outaouais, in summer
‘‘Maximum power point tracking based on sliding-mode control for
2019, to carry out a research.
output-series connected converters in photovoltaic systems,’’ IET Power
Electron., vol. 7, no. 4, pp. 914–923, Apr. 2014.
[16] Y. Zhou, C. N. M. Ho, and K. K.-M. Siu, ‘‘A fast PV MPPT scheme using
boundary control with second-order switching surface,’’ IEEE J. Photo-
volt., vol. 9, no. 3, pp. 849–857, May 2019.
[17] B. N. Alajmi, K. H. Ahmed, S. J. Finney, and B. W. Williams, ‘‘A max-
imum power point tracking technique for partially shaded photovoltaic
systems in microgrids,’’ IEEE Trans. Ind. Electron., vol. 60, no. 4,
pp. 1596–1606, Apr. 2013.
[18] H. Rezk, M. Aly, M. Al-Dhaifallah, and M. Shoyama, ‘‘Design and DHARAMPAL SINGH is currently pursuing the
hardware implementation of new adaptive fuzzy logic-based MPPT bachelor’s degree in electrical engineering with the
control method for photovoltaic applications,’’ IEEE Access, vol. 7, Punjab Engineering College, Chandigarh, India.
pp. 106427–106438, 2019. His main research interests include renewable
[19] V. R. Kota and M. N. Bhukya, ‘‘A novel global MPP tracking scheme energy and power electronics. He received the
based on shading pattern identification using artificial neural networks Mitacs Globalink Research Award to attend the
for photovoltaic power generation during partial shaded condition,’’ IET Université du Quebec en Outaouais, in summer
Renew. Power Gener., vol. 13, no. 10, pp. 1647–1659, Jul. 2019.
2019, to carry out a research.
[20] W.-M. Lin, C.-M. Hong, and C.-H. Chen, ‘‘Neural-network-based MPPT
control of a stand-alone hybrid power generation system,’’ IEEE Trans.
Power Electron., vol. 26, no. 12, pp. 3571–3581, Dec. 2011.
[21] H. Taheri, Z. Salam, K. Ishaque, and Syafaruddin, ‘‘A novel maximum
power point tracking control of photovoltaic system under partial and
rapidly fluctuating shadow conditions using differential evolution,’’ in
Proc. IEEE Symp. Ind. Electron. Appl. (ISIEA), Oct. 2010, pp. 82–87.
[22] S. Taheri, H. Taheri, Z. Salam, K. Ishaque, and H. Hemmatjou, ‘‘Modified
maximum power point tracking (MPPT) of grid-connected PV system
under partial shading conditions,’’ in Proc. 25th IEEE Can. Conf. Electr. SHAMSODIN TAHERI (Senior Member, IEEE)
Comput. Eng. (CCECE), Apr. 2012, pp. 1–4. received the B.Sc. degree in electrical engineer-
[23] S. Rajendran and H. Srinivasan, ‘‘Simplified accelerated particle swarm ing from the University of Mazandaran, Babolsar,
optimisation algorithm for efficient maximum power point tracking in Iran, in 2006, the master’s degree in electrical engi-
partially shaded photovoltaic systems,’’ IET Renew. Power Gener., vol. 10, neering from the Iran University of Science and
no. 9, pp. 1340–1347, Oct. 2016. Technology, Tehran, Iran, in 2009, and the Ph.D.
[24] K. Ishaque, Z. Salam, M. Amjad, and S. Mekhilef, ‘‘An improved particle degree in electrical engineering from the Univer-
swarm optimization (PSO)–based MPPT for PV with reduced steady-state
sité du Québec à Chicoutimi, Chicoutimi, QC,
oscillation,’’ IEEE Trans. Power Electron., vol. 27, no. 8, pp. 3627–3638,
Canada, in 2013. From 2013 to 2014, he worked
Aug. 2012.
[25] Y.-H. Liu, S.-C. Huang, J.-W. Huang, and W.-C. Liang, ‘‘A particle swarm with the Technical Services and Research Depart-
optimization-based maximum power point tracking algorithm for PV sys- ment, Saskpower, SK, Canada. In 2014, he joined the Université du Quebec
tems operating under partially shaded conditions,’’ IEEE Trans. Energy en Outaouais, Gatineau, QC, Canada. His main research interests include
Convers., vol. 27, no. 4, pp. 1027–1035, Dec. 2012. power systems, renewable energy, numerical modeling, and the integration
[26] K. Ishaque and Z. Salam, ‘‘A deterministic particle swarm optimization of electric vehicle into the grid.
maximum power point tracker for photovoltaic system under partial shad-
ing condition,’’ IEEE Trans. Ind. Electron., to be published.
[27] K. L. Lian, J. H. Jhang, and I. S. Tian, ‘‘A maximum power point track-
ing method based on Perturb-and-Observe combined with particle swarm
optimization,’’ IEEE J. Photovolt., vol. 4, no. 2, pp. 626–633, Mar. 2014.
[28] M. Li, W. Du, and F. Nian, ‘‘An adaptive particle swarm optimization
algorithm based on directed weighted complex network,’’ Math. Problems
Eng., vol. 2014, Apr. 2014, Art. no. 434972. DIEGO R. ESPINOZA-TREJO was born in San
[29] K.-H. Chao, ‘‘A high performance PSO-based global MPP tracker for a Luis Potosí, México. He received the B.S. degree
PV power generation system,’’ Energies, vol. 8, no. 7, pp. 6841–6858, in electronics engineering and the M.Sc. and Ph.D.
Jul. 2015. degrees in electrical engineering from the Uni-
[30] M. R. Islam, F. Rahman, and W. Xu, Advances in Solar Photovoltaic Power versidad Autónoma de San Luis Potosí, San Luis
Plants. Berlin, Germany: Springer, 2016.
Potosí, in 2001, 2004, and 2008, respectively.
[31] R. C. Eberhart and J. Kennedy, ‘‘A new optimizer using particle swarm
In 2009, he joined the Mechatronics Department,
theory,’’ in Proc. 6th Int. Symp. micro Mach. human Sci., vol. 1995, vol. 1,
pp. 39–43: New York, NY. Coordinación Académica Región Altiplano, Uni-
[32] R. Storn and K. Price, ‘‘Differential evolution—A simple and efficient versidad Autónoma de San Luis Potosí. He is cur-
heuristic for global optimization over continuous spaces,’’ J. Global rently a Professor with the Universidad Autónoma
Optim., vol. 11, no. 4, pp. 341–359, 1997. de San Luis Potosí and a National Researcher with the National Council of
[33] S. Das and A. Abraham, ‘‘Synergy of particle swarm optimization with Science and Technology (CONACYT). His main research interests include
evolutionary algorithms for intelligent search and optimization,’’ in Proc. active fault-tolerant control and fault diagnosis of power electronic systems,
IEEE Int. Congr. Evol. Comput., vol. 1, 2006, pp. 84–88. fault diagnosis of electric machines, and photovoltaic systems.

VOLUME 8, 2020 38491

 M. Joisher et al.: Hybrid Evolutionary-Based MPPT for PV Systems PSCs

EDRIS POURESMAEIL (Senior Member, IEEE) HAMED TAHERI received the B.Sc. degree
received the Ph.D. degree in electrical engineering in electrical engineering from the University of
from the Technical University of Catalonia (UPC- Mazandaran, Babolsar, Iran, in 2009, and the Ph.D.
Barcelona Tech), Barcelona, Spain, in 2012. After degree in electrical engineering from the ’Ecole
his Ph.D., he joined the University of Waterloo, de technologie supérieure, Montreal, QC, Canada,
Waterloo, Canada, as a Postdoctoral Research Fel- in 2017. From 2009 to 2011, he was a Research
low and then joined the University of Southern Assistant with the Inverter Quality Control Cen-
Denmark (SDU), Odense, Denmark, as an Asso- ter (IQCC), Universiti Teknologi Malaysia, Johor
ciate Professor. He is currently an Associate Pro- Bahru, Malaysia, which is responsible for testing
fessor with the Department of Electrical Engineer- PV inverters that are to be connected to the local
ing and Automation (EEA), Aalto University, Espoo, Finland. His main utility grid. From 2014 to 2018, he was with the Primax Technologies Inc.,
research activities focus on the application of power electronics in power Montreal, as a Power Electronics Designer for the UPS application. Since
and energy sectors. 2018, he has been with the Current, Powered by GE, Lachine, QC, as a
Power Electronics Designer for the application of LED driver for signage and
horticulture lighting systems. His current research interests include power
electronics, advanced control, and embedded programming.

38492 VOLUME 8, 2020