1 s2.0 S2772941922000060 Main

Download as pdf or txt
Download as pdf or txt
You are on page 1of 22

4 (2022) 200040

Contents lists available at ScienceDirect

Systems and Soft Computing


journal homepage: www.journals.elsevier.com/soft-computing-letters

A comparative study between deterministic and two meta-heuristic


algorithms for solar PV MPPT control under partial shading conditions
Arnold F. Sagonda *, Komla A. Folly
University of Cape Town, Electrical Engineering Department, Faculty of Engineering, Cape Town, South Africa

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

Keywords: During partial shading conditions (PSC), the PV modules in a solar PV array become reverse biased and act as a
Firefly algorithm (FA) load, resulting in hotspot issues. This can significantly decrease the efficiency of the solar PV. The general way to
Global maximum power point (GMPP) deal with PSC is to connect bypass diodes across the non-shaded PV module. However, this will alter the uniform
Partial shading conditions (PSC)
characteristics of the PV array, resulting in multiple power peaks i.e., a multi modal landscape. The conventional
Particle swarm optimization (PSO)
Perturb and observe (PnO)
gradient-based Perturb and Observe (PnO) algorithm generally used for maximum power point tracking (MPPT)
in solar PV is ineffective in finding the maximum power during PSC because it is prone to converge to local
optimum due to its nature of searching in a multimodal landscape. However, in this article an attempt is made to
show how the PnO behaves in a multimodal landscape considering different starting points. Robust stochastic
algorithms that are based on a population of search agents are used to guarantee convergence to the global MPP.
In this paper, the maximum power point under PSC was tracked using two meta-heuristic algorithms namely,
particle swarm optimization (PSO) and the firefly algorithm (FA). The performances of these algorithms in
tracking the maximum power point are evaluated. The efficiency, standard deviation (STD), and root mean
square error (RMSE) of the PSO and FA algorithms are compared to those of the PnO algorithm. The PSO was
found to have the lowest RMSE, and the FA had the lowest STD. The efficiency of the PSO and FA were relatively
the same. Simulation results show that PSO and FA-based MPPT algorithms can efficiently track the global
maximum power point (GMPP) irrespectively of the starting points. On the other hand, PnO is shown to be
unreliable under PSC even with different starting points.

1. Introduction maximum or minimum can be linear or nonlinear [8]. The objective


function can be single or multiple. Optimization algorithms that are used
The characteristic of PV array under partial shading (PS) comprises to solve the optimization problems can be classified mainly into two
multiple local maximum power points (LMPP) and one global maximum categories: deterministic approach and meta-heuristic approach [10].
power point (GMPP). Partial shading of PV modules connected in series When deterministic algorithms are run, only one solution is used at a
is known to have a significant impact on the capability of the solar PV to time. This solution will set the blueprint as the iteration continues [10].
deliver maximum energy, especially for utility-scale PV solar plants. Commonly used deterministic methods are the hill climbing (HC) and
Hotspot issues in PV arrays arise when shaded modules become reverse the PnO methods. If the algorithms are made to start at the same starting
biased and act as loads [1–6]. To mitigate this issue, bypass diodes are location, they will follow the same path whether they are run today or
used. The drawback of using bypass diodes is that they will alter the tomorrow [10]. Gradient-based deterministic algorithms use gradient
uniform PV characteristics curves of the array, resulting in multiple knowledge of the function to achieve optimization. Gradient based al­
power peaks [7]. The multiple power peaks can be seen as an optimi­ gorithms work very well for smooth unimodal problems. A good
zation problem. Optimization is a process of finding the best variable in example of such an algorithm is the Newton Raphson algorithm [8].
a function that will result in better outcomes amongst all the possible However, if there is some form of discontinuity in the objective function,
outcomes [8]. The region of decision variables is called the search space or for multimodal functions, the algorithm will not work properly [8]
[9]. The decision variables are subject to constraints so that a solution [10],. Another challenge for these types of algorithms is that they cannot
can be quickly obtained. The objective function to be optimized either cope with very large number of decision variables. Hence, heuristic and

* Corresponding author.
E-mail address: [email protected] (A.F. Sagonda).

https://doi.org/10.1016/j.sasc.2022.200040
Received 27 September 2021; Received in revised form 21 February 2022; Accepted 14 March 2022
Available online 16 March 2022
2772-9419/© 2022 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-
nc-nd/4.0/).
A.F. Sagonda and K.A. Folly Systems and Soft Computing 4 (2022) 200040

Fig. 1. (a) I-V and (b) P-V PV characteristics during PSC.

Fig. 2. PnO flow chart.

2
A.F. Sagonda and K.A. Folly Systems and Soft Computing 4 (2022) 200040

Fig. 3. Movements of firefly 1.

Fig. 4. Voltage control method.

meta-heuristic algorithms are designed to deal with this type of problem tradeoff between randomization and local search are the working
[9]. principles of all meta-heuristic algorithms [10]. Two major properties of
Meta-heuristic is a high level of searching using trial and error [8]. A any meta-heuristic algorithms are: intensification also known as

3
A.F. Sagonda and K.A. Folly Systems and Soft Computing 4 (2022) 200040

Fig. 5. Firefly flow chart for MPPT in solar PV systems.

exploitation and diversification, also known as exploration [8]. During clouds, etc.
diversification diverse solutions are created to search at a global level. There are two mitigation methods generally used to reduce the
During intensification focus is given on searching the local region by partial shading effect. The most complex and costly methods are hard­
taking advantage of the knowledge that a current best solution that has ware fixtures such as dynamic reconfiguration of solar PV modules ac­
been found in this region. The good combination of these two major cording to the shading patterns [14], micro inverter [18], DC power
properties of intensification and diversification will usually ensure that optimizer [19–23]. The less costly approach is to track the GMPP by
the best solution is found. Due to the landscape of multiple power peaks using artificial intelligence control method. This method is simpler than
under PS, the conventional gradient based MPPT algorithms such as PnO the above-mentioned mitigation methods and hence is the focus of this
[11] [12], hill-climbing (HC) [13], and incremental conductance (IC) paper. The paper also introduces how a gradient-based algorithm per­
[12] become inadequate because they cannot differentiate between forms in a multimodal landscape considering different starting points.
LMPP and the GMPP [14]. These algorithms are ideal for MPPT under Yaqoob Javed et al [24], showed that the PnO is getting stuck to local
uniform irradiance, where there is only a single MPP in the search space peaks although it was initialised at the same starting point for different
[11]. During PSC, the location of GMPP is not fixed, it may occur at PSC patterns. Theoretically, a gradient-based deterministic approach for
different voltage regions depending on solar insolation, cell temperature optimization should converge to a local peak which it finds but in this
and shading pattern [15]. Even though utility scale solar PV plants are study it has been shown that it can skip a LMPP. There is no formal study
purposefully located where PS is unlikely to occur, PSC is unavoidable in the literature that actually shows the behaviour of the PnO starting at
because of clouds and mutual shading of modules [16] [17]. For resi­ different points under PSC, to fill this gap this paper proposes different
dential solar systems, PSC is very common due to buildings, trees, starting points for PnO. GMPP based on meta-heuristic optimization

4
A.F. Sagonda and K.A. Folly Systems and Soft Computing 4 (2022) 200040

Fig. 6. Particle movement of PSO.

algorithms such as Grey Wolf (GW), Differential Evolution (DE), PSO, 3540 tons of waste could produce 1236.36 MWh of energy at a minimal
etc., have been proposed recently [1–5] [15]. However, the perfor­ processing cost of 165.22 $/ton. Chakraborty et al [28]. used the Harris
mances of many of these algorithms are heavily dependent on the Hawks Optimization (HHO) for optimal solar PV size and location on an
optimal selection of their control parameters. This optimal selection of IEEE 33-bus and IEEE 69-bus radial distribution system. The objective
control parameters is not a trivial task. Recently the FA which is a simple function was to minimize network power losses and improve the voltage
and robust algorithm has been proposed [8–10]. It has fewer control profile. The various decision variables including voltage, active and
parameters to tune compared to existing meta-heuristic optimization reactive power of the solar plant, phase angle at different buses were
algorithms such as PSO and has shown promising results. In this paper, under specific constraints. A penalty factor was introduced with the
the performance of meta-heuristic algoeithms such as FA and PSO for objective function for any overloaded line and voltage bus that would
MPPT in solar PV systems under PSC is investigated and is compared to violate the inequality constraints. The HHO was able to perform better
PnO which is a deterministic approach. The PnO is made to start at than the other algorithms and a power loss reduction of 64.42% for the
different points. The sensitivities of the control parameters of FA in 33-bus system and a 68.074% for the 69-bus system was achieved. Sabo
improving the performance of the algorithm are studied. The study et al [29], used the farmland fertility algorithm (FFA) for optimal tuning
shows that the starting point of the PnO in MPPT of PV systems under of power system stabilizers (PSS) control parameters to damp out low
PSC affects the power tracked and the search process is not predictable frequency oscillations in interconnected multi-machine power systems
as it was assumed in the previous studies. The well-tuned PSO and FA are during different operating conditions. The FFA was compared with the
shown to guarantee convergence to the GMPP. PSO and genetic algorithm (GA). The decision variables to be optimized
were the stabilizer gain and four time constants of the PSS lead-lag
2. The current research landscape on optimization structure. An eigenvalue-based objective function was used to find the
optimal PSS parameters on an IEEE 39- bus network. The objective was
Deterministic and Meta-heuristic approaches fall under the umbrella to minimize low-frequency oscillations by maximizing the damping
of soft computing. Soft computing has been applied to solve numerous ratio. FFA had better transient responses, faster convergence, and low
real-world problems. Recent applications of these techniques include computational cost than the other algorithms and thereby improved the
optimal PV model parameter estimation which was validated by network system dynamic stability.
experimental data [25]. The authors used a population-based eagle Mousavi et al [30], used four different multi-objective meta-heuristic
strategy gradient based optimizer with chaotic (ESCGBO) to estimate PV techniques to optimize two objectives, maximize the social impact and
static and dynamic model parameters. The ESCGBO was found to minimize the costs and air pollution of a blood supply chain network.
perform better than the other meta-heuristic techniques in finding the Special collecting vehicles (SCV) were used to gather blood from blood
5-parameter and 7-parameter single diode and double diode static PV transfusion centre and return back to the blood decomposition centre.
model respectively. The calculation of root mean square error was used The goal was to find optimal routes for the SCV with minimum trans­
to compare accuracy amongst the different algorithms and multiple runs portation costs whilst reducing carbon dioxide emissions and maxi­
(30) were used to check robustness through statistical analysis. Optimal mizing the collected blood that could be used. Normal distribution was
PV model parameter estimation was also achieved by [26]. Al-Refaie et used for the uncertainty of the amount of blood at transfusion centres
al [27], used a fuzzy optimization model and multi objectives for effi­ and the amount of blood sub-products. The findings of the paper were
cient municipal solid waste (MSW) management. The optimization that for maximum social impact, the numbers of SCV or their carrying
included maximization of transported quantities from dump sites and capacity should be increased which would inturn affect the route dis­
minimization of both greenhouse gas (GHG) emissions and transport tance, increase pollution emissions and costs. Therefore, a trade-off
costs. Some of the decision variables included waste quantity trans­ balance decision was required. Salehi-Amiri et al [31], used General
ported from each station, number of trips by trucks, generated revenues Algebraic Modeling System (GAMS) software and its exact optimizer
on a specific day…etc. The case study concluded that on a six-day period CPLEX to optimize a closed-loop avocado industry supply chain

5
A.F. Sagonda and K.A. Folly Systems and Soft Computing 4 (2022) 200040

Fig. 7. PSO flow chart for MPPT in solar PV.

network. Two objectives were to be optimized that is minimization of each centre, the optimum number of job employment, and the amount of
total costs and maximization of job employment at different stages of the stored product to meet demand. A sensitivity analysis was done on the
supply chain. The environmental impact was also considered by con­ effect of purchasing cost, changing the capacity at a facility, and
trolling the returned waste product. Mixed-integer linear programming changing demand. The conclusion of the study was that an increase in
(MILP) model was used. The model considered the optimum amount and product demand increased job opportunities drastically. Ghadami et al
location of facilities, the amount of avocados to be transported from [32], used Artificial Neural Network (ANN) time series to predict the

6
A.F. Sagonda and K.A. Folly Systems and Soft Computing 4 (2022) 200040

Fig. 8. The complete solar PV system in Simulink.

Table 1 Table 3
Parameters of Solar PV string. Irradiance (W/m2 ) exposed for each PS pattern.
PV String Parameter (STC) Total Value Units Pattern 1 (P1) Pattern 2 (P2) Pattern 3 (P3)

MPP power 1065.75 Watts 1000 400 300


MPP voltage 145 Volts 1000 400 300
MPP Current 7.35 Amps 800 600 600
Open circuit voltage 181.5 Volts 1200 800 800
Short circuit current 7.84 Amps 500 1000 1000

Fasihi et al [34], constructed a bi-objective multi-period model with


Table 2 binary variables to solve a fish closed-loop supply chain (CLSC). The
Design parameters used for Boost converter.
model considered forward and reverse logistics of the farmed fish. The
Parameters Value Units optimization required was the minimization of network total costs and
Duty cycle 51.6 % maximization of customer demand for both logistics. Three
Resistor 84.44 Ohms meta-heuristics methods were used to solve this complex model after
Inductor 0.0169 Henry realizing that exact methods did not give good solutions. Fathollahi-Fard
Capacitor 1.5273e− 5 Farads
et al [35], also used two hybrid meta-heuristics algorithms for a CLSC for
the entire industry considering uncertainties. Theophilus et al [36], used
energy consumption of a City. Then solar energy was forecasted using evolutionary algorithm (EA) to optimally solve a truck scheduling at a
PVSyst software to determine energy peak management. The goal of the cross-docking terminal (CDT). A mixed-integer model for the scheduling
paper was to encourage citizens to consume more energy during solar was formulated considering the rate of decay of perishable goods in the
peak energy times. The smart city management involved citizen presence of varying temperature of storage areas. The objective function
participation and motivation programs. D’Angelo et al [33], used ma­ was to minimize the total costs. EA was used to find optimal solutions for
chine learning techniques to differentiate between viral and bacterial the model. The EA was found to perform better than the exact CPLEX
meningitis. Formulas and rules were derived from known datasets. optimizer.

7
A.F. Sagonda and K.A. Folly Systems and Soft Computing 4 (2022) 200040

Fig. 9. Pattern 1: P-V curve Profile.

(ANNs) to find the MPP. Ahmed [45], reviewed the performances of the
Table 4
PSO, ACO, ANNs, Fuzzy-logic, etc., in improving MPPT under PS of solar
Testing results of multiple sets of N and T.
PV array. Dileep [46] reviewed deferent approaches based on an
N T optimum Worst Average number of Average improved PnO technique for designing MPPT for solar PV system under
value of max value of iterations to number of
PSC. Ling-Ling et al [37] used the improved gravitational search algo­
power (W) max power convergence max power
(W) (W) rithm for MPPT. The technique used time varying optimization param­
eters and it was shown to perform better than the conventional
5 40 745.4 744.5 8 744.4
20 40 746.6 744.6 10 745.5
gravitational search algorithm and PSO under PSC. Eltamaly et al [38]
35 40 746.7 744.7 12 745.5 used a novel scanning bat algorithm for MPPT during partially shaded
conditions. A direct control method was used where the duty cycle was
the variable. Kermadi et al [39] did an assessment of MPPT control
methods to access the advantages and disadvantages of using the direct
Table 5
or indirect control method. The PnO was used as the MPPT algorithm.
Tested sets of parameters (β0 = 1).
The effects of varying load and irradiance was used to test the two
Parameter Sets 1 2 3 4 5 methods. It was observed that the indirect method shows lower steady
α 0.01 0.10 0.25 0.40 1.00 state oscillations and less sensitive to rapid changes of solar irradiance
γ 0.01 0.10 0.40 0.60 1.00 and load. The drawback of the indirect method is the need for the
voltage controller which introduces additional circuitry as compared to
the direct method.
Table 6 Kermadi et al [41] did a critical review and performance evaluation
Performance of parameters based on convergence rate. on modern developments of MPPT techniques for solar systems during
PSC. The paper emphasises improved methods of conventional tech­
Parameter Sets 1 2 3 4 5
niques that include the 0.8 of open-circuit voltage (0.8 × Voc models)
Number of iterations (T) 40 40 40 40 40
and the recent skipping algorithms. Comparative performance of con­
Population (N) 20 20 20 20 20
Solar PV Power (W) 744.5 746.6 745.5 745.0 743.5 ventional, soft computing and hybrid methods were reviewed. Different
Objective function P1 algorithms from each category were tested and benchmarked under
specific operating PSC. The algorithms were evaluated based on the
ability to converge to GMPP, convergence time, voltage track and
2.1. Optimization in PV MPPT transient efficiency. The paper concluded that skipping algorithms like
the voltage window search (VWS) and search skip judge (SSJ) have a
Solar PV modules connected in series can undergo PS. During PSC, 100% global power convergence rate. The 0.8 ×Voc techniques which
the shaded module consumes the generated power produced by other include the modified incremental conductance (MIC) had a GMPP
non-shaded modules and dissipates heat [37–41]. Fig. 1. illustrates the convergence of 62%. The soft computing methods which included the
PSC showing the local and global power points. Flower pollination algorithm (FPA) had a GMPP convergence of 50%
In [41], the authors introduced a study of computational intelligence which was not anticipated but it was indicated that the importance of
techniques for MPPT, the advantages, and disadvantages of each tech­ optimizer parameter tuning is critical for GMPPT success. The hybrid
nique were reviewed. The control strategy of direct (duty cycle being techniques which include PSO-SSJ had a 93% GMPPT convergence rate.
controlled from the MPPT code) and indirect (other variables including Zaki Diab & Rezk [47] used a FPA for GMPPT and compared it with the
voltage being controlled from the MPPT code) was used to classify the PSO and differential evolution (DE). Statistical analysis was used to
approach of MPPT. Chao et al [42] used a modification of PSO for MPPT show that the FPA was better. Eltamaly et al [48] modified the PSO to
of solar PV module arrays that experience shading and failure. Fathabadi track the GMPP for varying PSC. Once the GMPP has been tracked the
[43] introduced a MPPT for a hybrid system consisting of solar PV, wind authors specified the importance of re-initialization to find the new
and fuel cell. The proposed MPPT does not need any expensive sensors as GMPP when the PS pattern changes. Wan et al [49] used a hybrid slap
it uses a unified algorithm to simultaneously track the MPPs of the swarm algorithm (SSA) with grey wolf optimizer (GWO) to track the
various energy sources. Kofinas et al [44] used artificial neural networks

8
A.F. Sagonda and K.A. Folly Systems and Soft Computing 4 (2022) 200040

Fig. 10. Convergence curve for the objective function.

would be that there is no communication between the agents. This


Table 7 method is then compared to well establish optimization methods like the
Meta-heuristic algorithm Parameters. PSO.
Parameters PSO FA

Population (N) 20 20 3. Implemented methods


Max iteration (T) 40 40
wmax 1 N/A 3.1. PnO
wmin 0.1 N/A
c1max 2 N/A
The PnO algorithm is a deterministic gradient-based algorithm. It
c1min 1 N/A
c2max 2 N/A moves along the solar P-V curve by small increments of the voltage and
c2min 1 N/A observing the resulting output power of the solar PV [11]. The PV
α N/A 0.1(P1&P3); 0.2 (P2) module voltage or the duty cycle can be used as the variable. Fig. 2
N/A 0.1 (P1); 1 (P2); 0.5 (P3)
γ
shows the flow chart of this algorithm. However, under PSC as shown in
Fig. 1, the algorithm can get stuck at anyone of the LMPP since it uses the
GMPP under static and dynamic PSC. The method was compared with gradient and has no intelligence or memory of knowing which peak is
PSO and PnO. It is shown that the hybrid method performed better in the GMPP or the LMPP. The starting point of the algorithm also plays an
terms of tracking efficiency and speed compared to the other methods. important role as the closest peak from where one starts is the likely
What is interesting in his paper is that the PnO skipped a peak under PSC peak point where the algorithm will converge according to literature.
for ten PV solar panels connected in series but the author did not Fig. 2 shows power as the objective function and whether current (P(k))
mention about it. Zhang et al [50] used an immune firefly algorithm and or previous (P(k-1)) is tracked by moving along the power curve using a
compared it to the regular firefly algorithm for MPPT and the immune constant step size (C). The step size is used to decide the direction of
firefly was faster at tracking the GMPP. Jamaludin et al [51] used SSA to movement of the variable voltage. For example, the algorithm will move
track the GMPP under dynamic PSC using direct duty cycle control. The in forward direction, if P(k) is greater than P(k-1) and the current V(k) is
SSA was modified in terms of having a duty cycle boundary (DCB). also greater than the previous voltage V(k-1). In this case, the voltage
Limiting the search space of probable solutions resulted in the SSA Vref should keep on increasing by a constant step size C. On the other
having fast convergence than other MPPT meta-heuristic methods. The hand, the algorithm will move in reverse direction, if P(k) is still greater
paper also shows the HC method skipping peaks under PSC but it was not than P(k-1) but V(k-1) is now greater than V(k). In this case, the voltage
clear where HC movement was initialized. Vref should decrease by C so as to track the power. It is well understood
that a trade-off between speed and accuracy is required in selecting the
step size value. Due to this search process, the PnO would clearly
2.2. Research gap
struggle to decide how to move in a multiple modal landscape.
The PnO is a deterministic approach that uses gradient information
for its search process to achieve optimization. In this article, an attempt 3.2. Firefly algorithm
to make the PnO have multiple starting points in a nonlinear objective
function can be considered as a stochastic approach to optimization [8]. The FA is a meta-heuristic algorithm [8] [10]. The flashing behav­
From the literature review, no formal approach has been suggested to iour of fireflies inspired this algorithm. The algorithm follows the in­
observe how the PnO would perform in a multimodal landscape verse square law i.e., light intensity (I) keeps on decreasing as the
considering different starting points. Most publications initialise the distance (r) increases, that is, I α r12 . Also, the air absorbs the light be­
variable of duty cycle at one starting point for the PnO and the tween the source and the object which further reduces the intensity of
assumption is that it would converge to the nearest peak it finds. This light [8, 10].
assumption has led to the PnO not being even used under PSC. In this
paper, voltage is used as a variable and it can be easily initiated at a 3.2.1. The idea of the FA algorithm
specific location along the objective function. The different starting
voltage points can be seen as different agents searching. The only issue • The fireflies are attracted to both males and females.

9
A.F. Sagonda and K.A. Folly Systems and Soft Computing 4 (2022) 200040

Fig. 11. MPPT by the PnO. (a) Different starting points of the PnO (b) Zoomed MPPT at around 0.02 s.

• Attractiveness and brightness are proportional to each other. That is


Is
to say, the less bright firefly will be drawn to the brighter firefly. The I(r) = (1)
r2
brightness is reduced as the distance increases.
• The location of a firefly on the objective function will determine its where Is is the source intensity and r is the distance.
brightness [8, 10]. For a medium with fixed light absorption coefficient γ, the light in­
tensity will change with distance r as shown in Eq. (2) [8].
3.2.2. Formulation of attractiveness from light intensity
To simplify the formulation of the variation in the light intensity and I = I0 e− γr
(2)
attractiveness of the FA, it follows that the attractiveness of a firefly is
where I0 is the original light intensity.
achieved based on its brightness [10]. Brightness is identified from the
In the Eq. (1), to avoid singularity at r = 0, the Gaussian form can be
objective function. The attractiveness (β) will vary with the distance ri
estimated by combining Eq. (2) and absorption to form Eq. (3) [8]:
between firefly i and firefly j and the amount of absorption of light in the
media. Eq. (1) shows how light intensity I(r) varies with distance [8]. I(r) = I0 e− γr2
(3)

10
A.F. Sagonda and K.A. Folly Systems and Soft Computing 4 (2022) 200040

Fig. 12. PSO and FA converging to GMPP. (a)MPPT during 0.03 s (b)MPPT zooming.

The intensity is proportional to attractiveness so the attractiveness of and j [8].


firefly is given by Eq. (4) [8]. √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
√ d
√∑ ( )2
β = β0 e− γr2
(4) rij =‖ xi − xj ‖= √ xi,k − xj,k (7)
k=1

where β0 is the attractiveness at r = 0.


A monotonically decreasing function can be used to represent the where xi,k is the kth component of the spatial coordinate xi of ith firefly,
attractiveness function β(r) as shown in Eq. (5) [8]. d is the dimension space. Eq. (7). can be written as Eq. (8) if the
dimension is 2.
β(r) = β0 e− γrm
, (m ≥ 1) (5) √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
( )2 ( )2̅
rij = xi − xj + yi − yj (8)
where m is a whole number greater than 1 when γ is fixed, the charac­
teristic length is Firefly i will be drawn to a brighter firefly j according to Eq. (9) [8].
2( )
Γ=γ − 1/m
1, m→∞ (6) xi = xi + β0 e− γrij xj − xi + α εi (9)
Eq. (7) shows the Cartesian distance that separates any two fireflies i

11
A.F. Sagonda and K.A. Folly Systems and Soft Computing 4 (2022) 200040

Fig. 13. P-V curve of pattern 2.

where the second term of Eq. (9) represents attraction. The last term in 3.3. PSO
Eq. (9) is randomization with α being the randomization parameter
controlling the step size, and εi is a vector of random numbers [10]. For PSO is a well-established stochastic optimization method [55]. It is a
most implementation, we can take β0 = 1. The trajectory of the FA is population-based algorithm meaning it uses a population of particles as
illustrated in Fig. 3. potential solutions. The PSO uses two simple principals for the move­
ment of particles to achieve global optimality, these are each particle
3.2.3. Parameters of firefly algorithm must move towards the best particle in the swarm Gbest , and move to­
To obtain optimum solutions, the FA parameters have to be tuned wards its own personal best Pbest [56]. The location of the particle xi is
adequately [8]. The FA has three parameters to tune namely, γ, α and β0 . updated by using Eq. (12) [56]. The movement of the particles is illus­
γ, represents how the attractiveness is going to change [52]. The trated in Fig. 6.
parameter α encourages diversification to allow global searching [8, 53,
xk+1 = xki + φk+1 (12)
54]. β0 usually works for a constant value of one for most applications i i

[8]. This simplifies the FA to only two tuneable parameters.


where φi is velocity. The velocity is calculated using Eq. (13)
{ } { }
3.2.4. FA in MPPT of solar PV φk+1
i = wφki + c1 r1 Pibest − xki + c2 r2 Gbest − xki (13)
For solar PV MPPT the FA uses the PV voltage as the variable and PV
power as the objective function to be maximized. where (w) is the inertia weight, r1 and r2 are random variables,c1 is the
The position (xi ) variable in Eq. (9) is selected as the voltage refer­ cognitive coefficient and c2 is the social coefficient. Pibest is the best
ence (Vref ). position of particle i, and Gbest is best location of the entire swarm [8,
FA is to find the best Vref voltage value that produces maximum 57].
power. Fig. 4 shows the whole system. This approach is known as the
voltage control method. Unlike the direct duty cycle control method, the 3.3.1. Parameters of PSO
advantage of this method is that the solar PV operating voltage is in­ w in Eq. (13) forces particles to follow the same direction of previous
dependent of the DC load variations, thus when the load changes, re- iteration [58]. c1 controls the particles attraction towards their own
initialization of the search process is not necessary [39]. The disad­ personal best [59]. c2 controls the attraction of particles towards the
vantage of this technique is the extra circuitry of the PI controller. current global of the swarm [60]. If c1<<c2 premature convergence to
The flow chart in Fig. 5 shows how to execute the FA for MPPT of swarm global best can occur and if c2<<c1 convergence might not occur
solar PV systems. First, a vector of Vref values is initialized. The initial­ as particles favour more their own personal best. Using parameters that
ized flies are ranked according to their power levels to obtain the vary with time will effectively tune the parameters for optimum global
GMP,PGbest . In the inner loop, the intensity of fly i is compared with fly j convergence [42] [58], [60, 61] [62],.
and note that j∕ = i. A linearly varying w over the generations is used. Eq. (14) shows the
If firefly power output Ppv,j>Ppv,i, the firefly position is updated by mathematical formulation.
Eq. (9). For MPPT in solar PV systems, d = 1 hence rij in Eq. (7) becomes
(iter)
[8] [10],. w(k) = wmax − (wmax − wmin ) × (14)
MaxIt
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
( )2 ( )
rij = xi − xj = xi − xj (10) where wmin and wmax are the lower and upper values of the inertia
weight, respectively, iter is the current iteration number and MaxIt is the
PGbest is updated by ranking the flies. To ensure GMPP convergence
maximum number of iterations [63]. Diversification is increased by
Eq. (11) is used.
increasing c1 and reducing c2 at the beginning of the search process [63].
| PGbest − Ppv,i | < ϵ i = 1…N (11) Intensification is encouraged at the end of optimization by increasing
c2 and reducing c1 [61]. Eq. (15) and Eq. (16) show the mathematical
where ϵ is the tolerance value. formulation [58].

12
A.F. Sagonda and K.A. Folly Systems and Soft Computing 4 (2022) 200040

Fig. 14. PnO converging to last peak from the right of the P-V pattern 2 profile. (a) Different starting points of the PnO (b) Zoomed MPPT at around 0.02 s.

iter present value.


c1 (k) = c1max − (c1max − c1min ) × (15)
MaxIt Other particle’s power values Ppv,i are ranked to see if the global
fitness value (gbest ) requires updating. Eq. (18) is used to check for GMP
iter convergence.
c2 (k) = c2min + (c2max − c2min ) × (16)
MaxIt
| Ppv,gbest − Ppv,i | < ϵ i = 1…n (18)
where c1min , c1max ,c2min and c2max are constants.
where ϵ is the tolerance value.
3.3.2. PSO in MPPT of solar PV Fig. 7 shows PSO flow chart for MPPT for a solar PV system.
The objective function is to extract maximum power from the solar
PV. The initial population will consist of a vector of voltage variables in 4. Partial shading patterns and tuning of the meta-heuristic
the range of where the optimum voltage solutions can be found. The control parameters
voltage control method shown in Fig. 4 is used again here. The particle
position xi is considered as Vref . The individual particle power Ppv,i is The complete system was built in Simulink as seen in Fig. 8. Five
calculated. Eq. (17) is used to check if Vref will produce a better personal solar PV modules were connected in series and the boost converter was
fitness value [59, 62]: designed for the MPPT. The parameters used in the design are shown
below in Tables 1 and 2.
Ppvi > Ppv , i − 1 (17)

If Eq. (17) is satisfied, Pbest,i is updated; otherwise, Pbest,i retains its

13
A.F. Sagonda and K.A. Folly Systems and Soft Computing 4 (2022) 200040

Fig. 15. MPPT by the PSO and FA. (a)MPPT during 0.03 s (b)MPPT zooming.

4.1. Partial shaded patterns The tuning of the control parameters to improve the convergence of the
FA is a challenging task. Since the parameters are problem dependant,
To test all the algorithms, three PSC patterns were used. The modules multiple test runs must be used to tune the control parameters for
were connected as shown in Fig. 8 . Table 3. shows the irradiance values different problems. It is well established that if N and T are set too small,
of each module for each PSC. The solar insulation values for a pattern are the tracking accuracy will be reduced and if they are set too large the
selected randomly so as to change the location of where the global tracking time will be increased [64]. Therefore, T was set to 40 (which is
power will occur. reasonable according to the literature) and multiple sets of N were
tested, each set of N was tested 20 times. Pattern 1 shown in Fig. 9 was
used to test the GMPPT with a theoretical value of 746.8 W.
4.2. Tuning of the control parameters of the meta-heuristic algorithm The results of the test are shown in Table 4. It can be seen from
Table 4 that when N = 5, the GMPP tracking accuracy was low, but the
4.2.1. Tuning of FA-MPPT control parameters convergence rate was fast due to the small size of population. For N =
The parameters of FA (α, γ and β0 ), the size of population (N) and the 20, the accuracy in tracking the GMPP value has increased compared to
maximum number of iteration (T) affect the performance of FA greatly.

14
A.F. Sagonda and K.A. Folly Systems and Soft Computing 4 (2022) 200040

Fig. 16. Exploration process of PSO and FA to find best Vref.

Fig. 17. Pattern 3: P-V curve Profile.

when N = 5 and the number of iterations needed to converge has also high value of α results in high step randomness. This resulted in the
increased. For N = 35, the improvement in accuracy compared to N = 20 particles moving away from the global optimum. Parameter set 2 was
is very negligible, but the convergence rate has slightly increased. selected for pattern 1 as it gave better PV power results. It should be
Therefore, N = 20 was used for the rest of the simulations. noted that for patterns 2 and 3, the control parameters are tuned again to
To investigate the sensitivities of the control parameters (α, γ) and obtain the optimal values. This is because each partial shading pattern
their impacts on the performance of the algorithm, five sets of control has a different landscape with a different location of the local and global
parameters were tested as given in Table 5. Note that β0 was kept con­ maximum points.
stant at unity in agreement with what is suggested in the literature. Fig. 10 shows the convergence curve of power against iterations. It is
Table 6 shows the performance of the algorithm for the different sets of observed that parameter set 2 requires less iteration to converge
the control parameters for pattern 1 (P1). It is observed that for the compared to the other parameter sets.
parameter set 1, small values of α and γ resulted in the fireflies As the values of α and, γ increased from parameter set 3 to parameter
converging quickly to a local optimum without adequate exploration. set 5, the number of iterations required to reach the maximum power
Parameter set 2 had a better trade-off between global search and local increased since the attraction becomes weaker as γ increases. The high
search and the algorithm converged to a global best value. As the values values of α and γ for the parameter set 5 compared to the other sets have
of α and, γ increased from set 3 to set 5, it can be observed that the al­ resulted in the particles moving away from the global optimum. How­
gorithm converged to a less accurate value compared to parameter set 2. ever, it should be mentioned that parameter set 5 would be ideal for
A high value of γ implies that the particle attraction becomes weak and a exploration during the beginning of the iterative process, and parameter

15
A.F. Sagonda and K.A. Folly Systems and Soft Computing 4 (2022) 200040

Fig. 18. MPPT of pattern 3 by the PnO. (a) Different starting points of the PnO (b) Zoomed MPPT at around 0.02 s.

set 1 would be ideal for exploitation at the end of the iterative process. nearest peak located to where it starts to operate as it moves along the
gradient slope. The PnO was made to start from different voltage points.
4.2.2. Tuning of PSO-MPPT control parameters These voltage points were purposefully selected to illustrate the local
The PSO parameter (w,c1and c2) selection was based on the guide­ search behaviour of the PnO as each point is near a different peak.
lines from the literature [65–68]. These parameters are time-varying Fig. 11(a) shows that the PnO produced different maximum power
and have limit range [67, 68]. Parameter w was made to decrease lin­ values when made to start at different voltage values.
early as the iterative process increases. A high value of w at the begin­ Fig. 11(b) shows the zoomed responses of circled portioning of
ning of the iteration would favour exploration then the value of w Fig. 11(a).
decreases linearly, for better exploitation at the end of the iterative Fig. 11 shows that when starting from 0 V and starting from 60 V the
process, c1 and c2 were tuned as discussed in the PSO parameter section. algorithm converged to LMP2 (645.4 W). However, when the algorithm
Since the parameters of PSO are time-varying, there was no need to alter starts at 60 V, the convergence is faster. When starting from 100 V, the
the limits of the parameters for different partial shading patterns as PnO converged to the GMPP (746.8 W) as this was the nearest peak to
compared to the constant FA control parameters which need to be tuned where it starts. However, when it starts at 140 V, the PnO converged to
for every pattern. LMP3 (618.5 W). These results are as expected since the PnO will
Table 7 illustrates the parameters selected for each algorithm. The converge to the first peak it detected.
parameters of the FA, α and γ were varied using a trial-and-error Fig. 12 shows MPPT by PSO and FA. Fig. 12.(a) shows the MPPT for
approach for every partial shading pattern. Each pattern (P1, P2 and 0.03 s Fig. 12(b) shows the zoomed responses of circled portion. The
P3) has its own parameters of α and γ. For the PnO, the step size of 1 V GMPP is easily found by PSO and FA when compared to PnO (see Fig. 12
was kept fixed for all simulations. However, the PnO algorithm starting (b)). However, close observation of Fig. 12(b) shows that the FA found a
point was varied to investigate whether it makes a difference in better solution than PSO. These techniques do not move along the
obtaining different results. gradient like the PnO does and therefore do not get stuck at LMP. They
use a population of voltage values and move according to the voltage
5. Simulation results value that gives off the best PV power.
Pattern 2 can be seen in Fig. 13. It has three local peaks and one
All the numerical data and design used for the simulations can be global peak.
found in [11] [15], [69], [70]. The PnO algorithm is a local search Fig. 14(a) shows the different starting points of the PnO and the
gradient-based method meaning that it is expected to converge to the convergence patterns. Fig. 14.(b) shows the zoomed results of the MPPT

16
A.F. Sagonda and K.A. Folly Systems and Soft Computing 4 (2022) 200040

Fig. 19. MPPT by the PSO and FA, for pattern 3. (a)MPPT during 0.03 s (b) MPPT zooming.

Table 8
Performance measurement of three different algorithm techniques.
ShadingPattern Technique GlobalPower, (W) BestPower, (W) WorstPower (W) AveragePower (W) STD RMSE Efficiency(%)

1 PSO 746.8 746.0 740.0 744.495 0.707 2.739 99.691


FA 746.8 746.7 744.0 745.495 0.494 1.940 99.825
PnO 746.8 645.0 645.0 645.0 0 101.8 86.368

2 PSO 476.5 475.0 420.0 468.875 17.677 15.068 98.399


FA 476.5 476.5 400.0 468.675 16.970 18.919 98.357
PnO 476.5 476.0 476.0 476.0 0 0.5 99.895

3 PSO 417.5 417.0 400.0 415.20 0.707 4.466 99.449


FA 417.5 416.5 390.5 414.25 0 6.416 99.221
PnO 417.5 361.0 361.0 361.0 0 56.5 86.467

by PnO. It can be seen that the algorithm converges to the last peak 120 V converged quickly to the GMPP which is its nearest peak. How­
which happens to be the GMPP (476.5 W) for all the starting points. ever, the fact that the starting points at 0 V and 40 V and 70 V skipped all
The PnO converged to the same GMPP despite starting at different the LMPs (LMP1-LMP3) of pattern 2 to converge to the GMPP was un­
locations. However, when the starting point is far from the GMPP, the expected. It is argued that PnO was able to track the GMPP from the
algorithm takes longer to converge. For example, the starting point at different starting voltages because of the patterns of irradiance (i.e.,

17
A.F. Sagonda and K.A. Folly Systems and Soft Computing 4 (2022) 200040

Table 9 Fig. 19 shows the MPPT of PSO and FA. Fig. 19(a) shows the MPPT
Averaged performance metrics vales over the three patterns. during 0.03 s and Fig. 19(b) shows the zoomed responses of circled
Algorithm RMSE STD Efficiency Computational Complexity portion in Fig. 19(a). The PSO and FA converged closer to the GMPP
with some ringing. FA seems a little bit far off than PSO as seen in Fig. 19
PSO 7.424 6.363 99.18 moderate
FA 9.092 5.821 99.13 moderate (b). The ringing occurs because the linear PI controller is unable to cope
PnO 52.933 0 90.91 low well with the nonlinearities.
To quantify the performance of the FA, PSO and PnO for the three PS
patterns, the following metrics are used. Standard deviation (STD), root
maximum peak is located at the highest voltage). It is believed that the mean square error (RMSE) and efficiency [47]. The metrics measure the
shape of the landscape of the solar profile played a significant role in values of the actual PV power (PPVa ) shown on each P-V curve profile
obtaining these results. Fig. 15(a) shows the MPPT of PSO and FA. and the estimated PV power (PPVe ) found by the algorithms. For the PnO
Fig. 15(a) shows the tracking during 0.03 s. Fig. 15(b) shows the only the values of these technique starting at zero is used as this gives a
zoomed responses of circled portion in Fig. 15(a). It can be seen from fair comparison for all PS patterns. The formulas of the performance
Fig. 15 that PSO and FA converged closer to the GMPP. However, FA is metrics are given bellow. The PSO and FA are executed 20 times. For the
seen to perform better than PSO in that it converges faster than PSO and PnO being a deterministic technique the values will never change after
gives a value closer to the GMPP than PSO with less ringing at steady being run multiple times. Table 8 shows the data of the performance of
state as seen in Fig. 15(b). the algorithms for the three PS patterns. Table 9 shows the averaged
Fig. 16 shows the exploration process of the FA and PSO for pattern 2 statistical comparison over the three patterns.
in obtaining the optimum voltage Vref value. Between 1.2x10− 3 sec. to √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
1.9 x 10− 3 sec both algorithms were exploring the search space before ∑nr ( )2
i=1 PPVe,i − PPVa
converging to the GMPP. However, FA converged slightly faster than RMSE = (19)
nr
PSO to a value of 156 V which is closer to the theoretical GMPP voltage
of 156.1 V. PSO converged to a value of 154 V which is slightly less than √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
∑n r ( )2
the theoretical value of GMPP. From Fig. 15(b) it can be noticed that the STD = i=1 PPVe,i − PPVe
(20)
PSO is oscillating between approximately 3.5x10− 3 sec and 4 x 10− 3 sec. nr
The FA voltage dip in Fig. 16 is significantly smaller compared to that of
PPVe,i
PSO. The PI controller parameters are slow to find the reference voltage Efficiency = × 100% (21)
PPVa
in Fig. 16 as they are not the best because they are linear controllers and
do not deal well with uncertainty of the set-point (exploration process). where nr is the number of executed runs of the simulation. i.e. 20 times
Fig. 17 shows the partial shading pattern 3. It has 3 LMPs and one From Table 9 it can be seen that the PSO has a lower RMSE compared
GMPP at 417.5 W. It can be seen that this pattern is similar in shape to to FA. FA has a lower STD than the PSO. The low STD represents stability
pattern 2 but the GMPP occurs at a different voltage point. Fig. 18(a) and robustness. The PSO has a STD of 6.363 and the FA 5.821. Using the
shows the results of the PnO from different starting voltage points. FA decreases the STD by 8.52% compared to the PSO. The PSO and the
Fig. 18(b) shows the zoomed responses of circled portion in Fig. 18(a). FA hve an RMSE of 7.424 and 9.092, respectively.. Employing the PSO
From Fig. 18 it can be seen that the PnO algorithm converged to LMP 3 decreases the RMSE by 18.34% compared to the FA. In terms of effi­
(361.5 W) which unfortunately is not the global peak. Again, when the ciency, the PSO and FA are relatively the same. The efficiency of the PnO
algorithm started at 0 V, 40 V and 70 V it skipped the closest peaks being initialized at zero still give a good value of 90.01%. This shows
which are LMP1 (171.4 W), LMP2 (345.9 W) and GMP (417.5 W) that even though the PnO can get stuck at a local power point during
respectively, and converges to the last peak LMP3 (361 W) which is a PSC, when averaged over the three different patterns it still can operate
local one. The shape of the landscape of the pattern was purposefully under PSC with good efficiency. The computational complexity of the
made to look similar to pattern 2 to illustrate the fact that PnO does not algorithms is also compared as seen in Table 9. The PnO is a simple
have the intelligence to differentiate between GMP and LMP. It seems technique that requires minimum computational effort compared to the
like the shape of the landscape of the pattern plays a role in how the PnO PSO or FA. The FA and PSO being applied to MPPT in solar PV are
behaves as the tacking of the peak is predetermined regardless of where moderately computationally demanding. The objective function is a
it starts. single nonlinear objective with only one variable being optimized. The

Fig. 20. Very rapidly varying solar irradiance profile curve.

18
A.F. Sagonda and K.A. Folly Systems and Soft Computing 4 (2022) 200040

Fig. 21. Solar PV output power under varying irradiance. (a) MPPT during the 0.2 s (b) MPPT at around 0.02 s.

population of voltage variables is bounded within a region where the PnO. Simulation results show that under PSC, PnO is unreliable for
MPP is expected to occur hence the PSO and FA converged close to the MPPT even for different starting points of the algorithm. The conver­
optimum value most of the time. All three algorithms require two sen­ gence of PnO to the GMPP depends on many factors including the
sors to compute the MPPT. One for voltage and the other for current. pattern of irradiance, the starting voltage etc. It was found that the only
The final test of a very rapidly changing solar insulation as shown in case where PnO was able to converge to the GMPP for all the different
Fig. 20 was used to further verify the performances of the algorithms. starting points is when the GMPP is the last peak and is located near the
There is one GMPP under these conditions and the PnO is expected to highest voltage. It has been shown that generally, PnO converges
perform better. The results are shown in Fig. 21. Fig. 21(a) shows the directly to the LMPP right next to its starting point or it might skip the
MPPT of PSO, FA and PnO. Fig. 21(b) shows the zoomed responses of LMPP and converge to another LMPP. Under PSC, the GMPP can exist at
circled portion in Fig. 21(a). It can be seen that FA and PSO tracked the any point; therefore, it is hard to know where to initialize the starting
irradiance with great accuracy and very minimum oscillations as seen in point for the PnO. Even if the pattern is known, the simulation results
Fig. 21(b). The PnO algorithm, on the other hand, was not able to track indicate that the PnO might still converge to LMPP even though it
the rapid change of irradiance with good accuracy. There are significant started right next to the GMPP as shown in pattern 3. The condition
errors as can be seen in Fig. 21(b). under which the PnO is able to skip the GMPP is still unclear and needs
further investigations.
6. Conclusion The meta-heuristic algorithms PSO and FA, on the other hand, have
intelligence and are able to easily track the GMPP. These algorithms can
The performances of FA in MPPT in solar PV system under PSC were find the best voltage value that gives the GMP from a population that is
investigated. To assess the efficiency of FA it was compared to PSO and initialized randomly in the search space. The ringing that is occurring at

19
A.F. Sagonda and K.A. Folly Systems and Soft Computing 4 (2022) 200040

steady state for the FA and PSO is due to the fact that the PI controller is • The PnO results are limited to only simulation. Hardware imple­
linear and cannot cope well with the high nonlinearity of solar PV and mentation will be done in the future to see if similar results are
the boost converter. obtained.
In terms of the efficiency of the algorithms, it can be said that PSO • For the statistical comparison, more sample sizes in terms of PS
and FA, have relatively the same efficiency. However, these algorithms patterns are required to further quantify the metrics
are more efficient than PnO. This is because the PSO and FA converge • In the model used in this manuscript, only five power peaks could be
closer to the GMPP most of the time whilst the PnO can skip peaks and created. In the future more PV modules could be connected in series
converge to a LMPP. to create more power peaks and similar tests can be done.
Under rapid varying irradiance, the PSO and FA achieved MPPT with
great accuracy compared to the PnO algorithm. It should be noted that Declaration of Competing Interest
FA control parameters in this paper were fixed thus they have to be
tuned for every pattern to obtain good results. It is expected that self- The authors declare that they have no known competing financial
adaptive FA will be used in our future work. This would enable the interests or personal relationships that could have appeared to influence
control parameters of FA to be adaptive and be tuned automatically for the work reported in this paper.
different partial shaded patterns. Some other limitations and future
research include the following. Acknowledgement

• It is still not clear why the PnO skips a local peak. More PS patterns This work is based on research supported in part by the NRF UID
need to be tested to observe what causes this. 118550.
• The paper was limited to static PSC and in the future dynamic PSC
will be tested.

Appendix

Abbreviations

Particle Swarm Optimization


PSO
FA Firefly Algorithm
PnO Perturb and observe
PV Photovoltaic
GMPP Global maximum power point
LMPP Local maximum power point
LMP Local maximum power
MPPT Maximum power point tracking
MPP Maximum power point
PS Partial shading
PSC Partial shading conditions
HC Hill climbing
IC Incremental conductance
DC Direct current
GW Grey Wolf
DE Differential Evolution
ESCGBO Eagle strategy gradient based optimizer with chaotic
MSW municipal solid waste
GHG greenhouse gas
HHO Harris Hawks Optimization
IEEE Institute of Electrical and Electronics Engineers
FFA farmland fertility algorithm
PSS Power Systems Stabilizers
GA Genetic algorithm
SCV Special collecting vehicles
MILP mixed-integer linear programming
ANN Artificial Neural Network
CLSC Closed loop supply chain
EA Evolutionary algorithm
CDT Cross-docking terminal
ACO Ant Colony optimization
VWS voltage window search
SSJ search skip judge
MIC modified incremental conductance
FPA Flower pollination algorithm
SSA slap swarm algorithm
N/A Note applicable
STC Standard test conditions
R distance
I Light intensity
В attractiveness
Γ light absorption coefficient
D dimension space
(continued on next page)

20
A.F. Sagonda and K.A. Folly Systems and Soft Computing 4 (2022) 200040

(continued )
εi vector of random numbers
α Step size
PI Proportional Integrator
W inertia weight
c1 cognitive coefficient
c2 social coefficient
r1 random variables
MaxIt maximum number of iterations
Iter current iteration number
T number of iteration
N size of population
P1 Pattern 1
P2 Pattern 2
P3 Pattern 3
W Watts
RMSE root mean square error
STD Standard deviation
nr number of executed runs of the simulation
DCB duty cycle boundary

References [20] R. Bell, R.C.N. Pilawa-Podgurski, Decoupled and distributed maximum power
point tracking of series-connected photovoltaic submodules using differential
power processing, IEEE J Emerg Sel Top Power Electron 3 (2015) 881–891.
[1] S. Titri, C. Larbesb, K.Y. Toumi, K. Benatchba, A new MPPT controller based on the
[21] L.F.L. Villa, T.P. Ho, J.C. Crebier, B Raison, A power electronics equalizer
Ant colony optimization algorithm for Photovoltaic systems under partial shading
application for partially shaded photovoltaic modules, IEEE Trans. Ind. Electron.
conditions, Appl. Soft Comput. 58 (2017) 465–479.
60 (2013) 1179–1190.
[2] Benyoucef As, A. Chouder, K. Karaa, S. Silvestre, Ait sahed O,. Artificial bee colony
[22] KukitaA. UnoM, Single-switch voltage equalizer using multistacked buck–boost
based algorithm for maximum power point tracking (MPPT) for PV systems
converters for partially shaded photovoltaic module, IEEE Trans. Power Electron.
operating under partial shaded conditions, Appl. Soft Comput. 32 (2015) 38–48.
30 (2015) 3091–3105.
[3] S. Mohanty, B. Subudhi, P.K. Ray, A New MPPT Design Using Grey Wolf
[23] E. Koutroulis, F. Blaabjerg, A new technique for tracking the globalmaximum
Optimization Technique for Photovoltaic System under Partial Shading Conditions,
power point of PV arrays operating under partial-shading conditions, IEEE Journal
EEE Trans Sustain Energy 7 (2016) 181–188.
of Photovolttaics 2 (2012) 184–190.
[4] M. Kermadi, E.M. Berkouk, Artificial intelligence-based maximum power point
[24] M. Yaqoob Javed, F. Murtaza, Q. Ling, S. Qamar, M. Majid Gulzar, A nocel MPPT
tracking controllers for Photovoltaic systems: comparative study, Renew Sustain
design using generalized pattern search for partial shading, Energy Build 133
Energy Rev 69 (2017) 369–386.
(2016) 59–69.
[5] J. Ahmed, Z. Salam, A Maximum Power Point Tracking (MPPT) for PV system using
[25] A. Ramadan, S. Kamel, M.H. Hassan, T.M. Véliz, A.M. Eltamaly, Parameter
Cuckoo Search with partial shading capability, Appl. Energy 119 (2014) 118–130.
Estimation of Static/Dynamic Photovoltaic Models Using a Developed Version of
[6] A. Gupta, Y.K. Chauhan, R.K. Pachauri, A comparative investigation of maximum
Eagle Strategy Gradient-Based Optimizer, Sustainability 13 (2021) 13053.
power point tracking methods for solar PV system, Sol. Energy 136 (2016)
[26] D. Yousri, A. Fathy, H. Rezk, T.S. Babu, M.R. Berber, A reliable approach for
236–253.
modeling the photovoltaic system under partial shading conditions using three
[7] Spagnuolo G Romero-CadavalE, L.G. Franquelo, Suntio T Ramos-PajaCA, W.
diode model and hybrid marine predators-slime mould algorithm, Energy
M Xiao, Grid-connected photovoltaic generation plants: components and operation,
Conversion and Management 243 (2021), 114269.
IEEE Ind. Electron. Mag. 7 (2013) 6–20.
[27] A. Al-Refaie, A. Al-Hawadi, N. Lepkova, A fuzzy optimization model for methane
[8] X.S Yang, Engineering Optimization An Introduction with Metaheuristic
gas production from municipal solid waste, Soft Computing Letters 3 (2021),
Applications, A JOHN WILEY & SONS, INC, Cambridge, United Kingdom, 2010.
100019.
[9] A. Ariyaratne, T.G.I. Fernando, A comparative Study on Nature Inspired
[28] S. Chakraborty, S. Verma, A. Salgotra, M.R. Elavarasan, D. Elangovan, L. Mihet-
Algorithms with Firefly Algorithm, Int. Journal of Eng. and Technology. 4 (2014)
Popa, Solar-Based DG Allocation Using Harris Hawks Optimization While
611–617.
Considering Practical Aspects, Energies 14 (2021) 5206.
[10] Yang X. Nature-Inspired Metaheuristic Algorithms. 2nd Ed. Cambridge, United
[29] A. Sabo, N.I. Abdul Wahab, M.L. Othman, M.Z.A. Mohd Jaffar, H. Beiranvand,
Kingdom, 2010.
Optimal design of power system stabilizer formultimachine power system using
[11] A. Sagonda, K. Folly, P. Kenneth Ainah, Comparison of Three Techniques for
farmland fertility algorithm, Int TransElectr Energ Syst (2020) e12657.
Maximum Power Point Tracking of Solar PV, IEEE Intenaional confrence on fuzzy
[30] R. Mousavi, A. Salehi-Amiri, A. Zahedi, M. Hajiaghaei-Keshteli, Designing a supply
systems (2018).
chain network for blood decomposition by utilizing social and environmental
[12] R. Pilawa-Podgurski, W. Li, I. Celanovic, D.J. Perreault, Integrated CMOS energy
factor, Computers & Industrial Engineering 160 (2021), 107501.
harvesting converter with digital maximum power point tracking for a portable
[31] A. Salehi-Amiri, A. Zahedi, F. Gholian-Jouybari, Z.E. Rodríguez Calvo,
thermos photovoltaic power generator, IEEE J Emerg Sel Top Power Electron 3
M Hajiaghaei-Keshteli, Designing a Closed-loop Supply Chain Network Considering
(2015) 1021–1035.
Social Factors; A Case Study on Avocado Industry, Appl Math Model 101 (2022)
[13] B.N. Alajmi, K.H. Ahmed, S.J. Finney, B.W. Williams, Fuzzy logic-control approach
600–631.
of a modified hill-climbing method for maximum power point in micro grid
[32] N. Ghadami, M. Gheibi, Z. Kian, G.M. Faramarz, R. Naghedi, M. Eftekhari, A.
standalone photovoltaic system, IEEE TransPower Electron n.d 26 (2011)
M. Fathollahi-Fard, A.M. Dulebenets, G. Tian, Implementation of solar energy in
1022–1030.
smart cities using anintegration of artificial neural network, photovoltaic system
[14] N. Femia, G. Lisi, G. Petrone, G. Spagnuolo, M. Vitelli, Distributed maximum power
and classical Delphi methods, SustainableCities and Society (2021).
point tracking of photovoltaic arrays: novel approach and system analysis, IEEE
[33] G. D’Angelo, R. Pilla, C. Tascini, S Rampone, A proposal for distinguishing between
Trans. Ind. Electron. 55 (2008) 2610–2621.
bacterial and viral meningitis using genetic programming and decision trees, Soft
[15] A. Sagonda, K. Folly, Maximum Power Point Tracking in Solar PV Under Partial
comput 23 (2019) 11775–11791.
Shading Conditions Using Stochastic Optimization Techniques, IEEE Congress on
[34] Fasihi M., Tavakkoli-Moghaddam R., Najaf E.S., Hajiaghaei-Keshteli M. Optimizing
evolutionary computation (2019).
a bi-objective multi-period fish closed-loop supply chain network design by three
[16] T. Kerekes, E. Koutroulis, D. Sera, R. Teodorescu, M. Katsanevakis, An optimization
multi-objective meta-heuristic algorithm. 2022.
method for designing large PV plants, IEEEJournal Photovoltaics 3 (2013)
[35] A.M. Fathollahi-Fard, M.A. Dulebenets, M. Hajiaghaei–Keshteli, R. Tavakkoli-
814–822.
Moghaddam, M. Safaeian, H. Mirzahosseinia, Two hybrid meta-heuristic
[17] Lumby A.M. and B. Utility Scale Solar Power Plants. A Guid Dev Investors World
algorithms for a dual-channel closed-loop supply chain network design problem in
Bank 2012.
the tire industry under uncertainty, Advanced Engineering Informatics (2022).
[18] G. Velasco-Quesada, F. Guinjoan-Gispert, R. Pique-Lopez, M. Roman-Lumbreras,
[36] O. Theophilus, M.A. Dulebenets, J. Pasha, Y. Lau, A.M. Fathollahi-Fard,
A. Conesa-Roca, Electrical PV array reconfiguration strategy for energy extraction
A. Mazaheri, Truck Scheduling Optimization at a Cold-Chain Cross-Docking
improvement in grid connected PV systems, IEEE Trans. Ind. Electron. 56 (2009)
Terminal with Product Perishability Consideration, Computers & Industrial
4319–4331.
Engineering (2022).
[19] D. Nguyen, B. Lehman, An adaptive solar photovoltaic array using model-based
[37] L. Ling-Ling, L. Guo-Qian, T. Ming-Lang, T. Kimhua, L Ming, A maximum power
reconfiguration algorithm, IEEE Trans. Ind. Electron. 55 (2008) 2644–2654.
point tracking methon for PV system with improved gravitational search
algorithm, Appl Soft Comput 65 (2018) 333–348.

21
A.F. Sagonda and K.A. Folly Systems and Soft Computing 4 (2022) 200040

[38] A. Eltamaly, M.S. Al-Saud, A.G. Abokhalil, A novel scanning bat algorithm strategy [53] S. Arora, S. Singh, The Firefly Optimization Algorithm:convergence Analysis and
for maximum power pointtracker of partially shaded photovoltaic energy systems, Parameter Selection, Int. J. Comput. Appl. 69 (2013) 48.
Ain Shams Engineering Journal (2020). [54] K. Sundareswaran, S.P.S Peddapati, MPPT of PV systems under partial shaded
[39] M. Kermadi, S. Mekhilef, Z. Salam, J. Ahmed, E.M. Berkouk, Assessment of conditions through a colony of flashing fireflies, IEEE Trans. Energy Convers. 29
Maximum Power Point Trackersperformance Using Direct and Indirect Control (2014) 463–472.
Methods, Wiley, 2020. [55] K. Sundareswaran, V.V. Kumar, S. Palani, Application of a combined particle
[40] A.M. Eltamaly, M.S. Al-Saud, A.G. Abokhalil, M.H. Farh, Photovoltaic maximum swarm optimization and perturb and observe method for MPPT in PV systems
power pointtracking under dynamic partial shadingchanges by novel adaptive under partial shading conditions, Renew Energy 75 (2015) 308–317.
particle swarm optimization strategy, Transactions of the Institute ofMeasurement [56] Y.-.H. Liu, S.-.C. Huang, J.-.W. Huang, W.-.C. Liang, A Particle Swarm
and Control (2019). Optimization-Based Maximum Power Point Tracking Algorithm for PV Systems
[41] M. Kermadi, Z. Salam, A.M. Eltamaly, J. Ahmed, S. Mekhilef, C. Larbes, E. Operating Under Partially Shaded Conditions, IEEE Trans. Energy Convers. 27
M. Berkouk, Recent Developments of MPPT Techniques for PV Systems under (2012) 1027–1035.
Partial Shading Conditions: A Critical Review and Performance Evaluation, IET [57] M. Clerc, J. Kennedy, The particle swarm—Explosion, stability, and convergence in
Renewable Power Generation, 2020. a multi-dimensional complex space, IEEE Trans. Evol. Comput. 6 (2002) 58–73.
[42] K.H. Chao, Y.S. Lin, U.D. Lai, Improved particle swarm optimization for maximum [58] P.N. Suganthan, Particle swarm optimizer with neighborhood operator, IEEE Proc.
power point tracking in photovoltaic module arrays, Appl. Energy 158 (2015) of the 1999 Congr on Evol Comput CEC99 3 (1999) 1958–1962.
609–618. [59] A. Ratnaweera, S.K. Halgamuge, H.C. Watson, Self-Organizing Hierarchical
[43] H. Fathabadi, Novel highly accurate universal maximum power point tracker for Particle Swarm Optimizer With Time-VaryingAcceleration Coefficients, IEEE
maximum power extraction from hybrid fuel cell/photovoltaic/wind power Trans. Evol. Comput. 8 (2004) 240–255.
generation systems, Energy 116 (2016) 402–416. [60] Y. Shi, R.C. Eberhart, Empirical study of particle swarm optimization, IEEE Proc.
[44] Kofinas P., Dounis A.I., Papadakis G.A.M. An Intelligent MPPT controller based on on Congress on Evol Comput 3 (1999) 1945–1950.
direct neural control for partially shaded PV system. Energy Build.2915;90:51–64. [61] J. Kennedy, R. Eberhart, Particle swarm optimization, IEEE Int. Conf. Neural
[45] J.S.Z Ahmed, A critical evaluation on maximum power point tracking methods for Networks (1995) 1942–1948.
partial shading in PV systems, Renew Sustain Energy Rev 47 (2015) 933–953. [62] K.L. Lian, J.H. Jhang, I.S Tian, A Maximum Power Point Tracking Method Based on
[46] G.S.S Dileep, Maximum power point tracking of solar photovoltaic system using Perturb-and-Observe Combined With Particle Swarm Optimization, IEE J of
modified perturbation and observation method, Renew Sustain Energy Rev 50 Photovoltaic 4 (2014) 626–633.
(2015) 109–129. [63] Z. Li-ping, Y. Huan-jun, H. Shang-xu, Optimal choice of parameters for particle
[47] A.A. Zaki Diab, H. Rezk, Global MPPT based on flower pollination and differential swarm optimization, J. Zhejiang Univ.-Sci A 6 (2005) 528–534.
evolutionalgorithms to mitigate partial shading in building integrated PV system, [64] B. MOY-, Y.-.Z. Ma, Q.-.Y. Zheng, Optimal Choice of Parameters for Firefly
Solar Energy 157 (2017) 171–186. Algorithm, Fourth Int. Conf. Digit. Manuf. Autom. 06 (2013) 887–892.
[48] A.M. Eltamaly, H.M. Farh, M.S. Al Saud, Impact of PSO Reinitialization on the [65] Y. Dai, L. Liu, Y. Li, An Intelligent Parameter Selection Method for Particle Swarm
Accuracyof Dynamic Global Maximum Power Detection of Variant Partially Optimization Algorithm, Fourth Int. Jt. Conf. Comput. Sci. Optim. 04 (2011)
Shaded PV Systems, Sustainability 11 (2019) 2091. 960–964.
[49] Y. Wan, M. Mao, L. Zhou, Q. Zhang, X. Xi, C. Zheng, A Novel Nature-Inspired [66] F. Zhang, J. Li, C. Feng, Y. Wu, In-depth investigation of effects of partial shading
Maximum Power PointTracking (MPPT) Controller Based on SSA-GWO Algorithm on PV array characteristics, IEEE Power Eng 09 (2012) 1–4.
for Partially Shaded Photovoltaic Systems, Electronics (Basel) 8 (2019) 680. [67] M.C. Chih, C.-.J. Lin, M.-.S. Chern, T.-.Y. Ou, Particle swarm optimization with
[50] M. Zhang, Z. Chen, L. Wei, An Immune Firefly Algorithm for Tracking time-varying acceleration coefficients for the multi-dimensional knapsack problem,
theMaximum Power Point of PV Array under PartialShading Conditions, Energies Appl Math Model 38 (2014) 1338–1350.
12 (2019) 3083. [68] A.R. Jordehi, Time varying acceleration coeffcients particle swarm optimisation:a
[51] M.N.J. Jamaludin, F.M. NAIM Naim Tajuddin, J. Ahmed, A. Azmi, A.S. Azmi, H. new optimisation algorthm for estimating parameters of PV cells and modules,
N. Ghazali, S.T. Babu, H.H Alhelou, An Effective Salp Swarm Based MPPT Energy Conversion and Management 129 (2016) 262–274.
forPhotovoltaic Systems Under Dynamicand Partial Shading Conditions, Ieee [69] A.F. Sagonda, K. Folly, Modelling and Tracking of the Global Maximum Power
Access (2021) 9. Point in Shaded Solar PV Systems Using Computational Intelligence, University of
[52] D.F. Teshome, C.H. Lee, Y.W. Lin, K.L. Lian, A Modified Firefly Algorithm for Cape Town, Cape Town, 2019. https://open.uct.ac.za/handle/11427/30973.
Photovoltaic Maximum Power Point Tracking Control Under Partial Shading, IEEE [70] P.K. Ainah, K.A. Folly, A.F Sagonda, Increasing PV penetration With D-STATCOM in
J Emerg Sel Top Power Electron 5 (2017). a Radial LV Distribution Feeder, Southern Africa Universities Power Engineering
Conference, 2018.

22

You might also like