Renewable and Sustainable Energy Reviews - UMEXPERT · PDF fileState of the art artiﬁcial...

Renewable and Sustainable Energy Reviews 64 (2016) 435–455

Contents lists available at ScienceDirect

Renewable and Sustainable Energy Reviews

http://d1364-03

n CorrE-m

journal homepage: www.elsevier.com/locate/rser

State of the art artificial intelligence-based MPPT techniques formitigating partial shading effects on PV systems – A review

M. Seyedmahmoudian a,n, B. Horan a, T. Kok Soon b, R. Rahmani c, A. Muang Than Oo a,S. Mekhilef d, A. Stojcevski e

a School of Engineering, Deakin University, Victoria, Australiab Department of Computer System and Technology, Faculty of Computer Science and Information Technology, University of Malaya, Malaysiac School of Software and Electrical Engineering, Swinburne University of Technology, Australiad Department of Electrical Engineering, University of Malaya, Malaysiae Centre of Technology, RMIT University, Vietnam

a r t i c l e i n f o

Article history:Received 31 August 2015Received in revised form29 May 2016Accepted 26 June 2016Available online 5 July 2016

Keywords:Maximum power point trackingPhtovoltaic systemsPartial shadingArtificial intelligenceSoft computing

x.doi.org/10.1016/j.rser.2016.06.05321/& 2016 Elsevier Ltd. All rights reserved.

esponding author.ail address: [email protected] (M. Sey

a b s t r a c t

Given the considerable recent attention to distributed power generation and interest in sustainableenergy, the integration of photovoltaic (PV) systems to grid-connected or isolated microgrids has becomewidespread. In order to maximize power output of PV system extensive research into control strategiesfor maximum power point tracking (MPPT) methods has been conducted. According to the robust, re-liable, and fast performance of artificial intelligence-based MPPT methods, these approaches have beenapplied recently to various systems under different conditions. Given the diversity of recent advances toMPPT approaches a review focusing on the performance and reliability of these methods under diverseconditions is required. This paper reviews AI-based techniques proven to be effective and feasible toimplement and very common in literature for MPPT, including their limitations and advantages. In orderto support researchers in application of the reviewed techniques this study is not limited to reviewingthe performance of recently adopted methods, rather discusses the background theory, application toMPPT systems, and important references relating to each method. It is envisioned that this review can bea valuable resource for researchers and engineers working with PV-based power systems to be able toaccess the basic theory behind each method, select the appropriate method according to project re-quirements, and implement MPPT systems to fulfill project objectives.

& 2016 Elsevier Ltd. All rights reserved.

Contents

1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4362. Photovoltaic system characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 437

2.1. Normal conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4372.2. Partial shading conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 437

3. Common SC based MPPT techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4383.1. Artificial neural network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 438

3.1.1. Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4383.1.2. ANN based MPPT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 438

3.2. Fuzzy logic control. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 440
3.2.1. Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4403.2.2. FLC-based MPPT. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 441
3.3. Particle swarm optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 441
3.3.1. Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4413.3.2. PSO-based MPPT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4423.3.3. Discussion and previous studies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 442
edmahmoudian).

www.sciencedirect.com/science/journal/13640321

www.elsevier.com/locate/rser

http://dx.doi.org/10.1016/j.rser.2016.06.053



http://crossmark.crossref.org/dialog/?doi=10.1016/j.rser.2016.06.053&domain=pdf



mailto:[email protected]


M. Seyedmahmoudian et al. / Renewable and Sustainable Energy Reviews 64 (2016) 435–455436

3.4. Ant colony optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 444
3.4.1. Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4443.4.2. ACO-based MPPT techniques. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 444
3.5. Genetic algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 445
3.5.1. Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4453.5.2. GA-based MPPT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 446
3.6. Differential evolution. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 447
3.6.1. Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4473.6.2. Initialization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4473.6.3. Mutation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4473.6.4. Crossover . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4473.6.5. Selection. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 448
3.7. DE-based MPPT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4483.8. Emerging methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 448

3.8.1. Radial movement optimization (RMO) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4483.8.2. Cuckoo search method (CS). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4493.8.3. Grey wolf optimization (GWO) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4503.8.4. Firefly optimization algorithm (FOA) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 450

4. Analytical comparison . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4505. Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 452References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 452

1. Introduction

The demand for photovoltaic (PV) power generation in powersystems and the distribution sector is growing significantly. Re-search shows that the contribution of PV systems to energy gen-eration was approximately 14,000 MW in 2010 and is expected tobe 70,000 MW in 2020 [1]. Australia is a leader in utilizing solarenergy resources [2], and had a solar power generation capacity of115 MW in 2009, which contributed an estimated 0.1–0.2% of thetotal electricity production. Given the hot, dry, and sunny climate,ideal for solar energy utilization, there is a projected target for 20%of total electricity supply to derive from renewable energy by 2020[3]. Despite recent advancements in PV utilization-related factors,such as reduction in cost, cell efficiency increases, and improvedstructural integration to buildings [1], the low energy conversionefficiency of PV systems remains a major impediment to utilizationof PV power generation and being able to accurately achievemaximum power point tracking (MPPT) is critically important.Another challenge with PV power generation is the heavy de-pendence on environmental factors, including solar irradiance andambient temperature. Therefore, the control unit must be com-piled through a capable MPPT technique to harvest the maximumenergy from the output terminal of installed PV arrays by pro-viding an appropriate duty cycle to operate the embedded DC–DCconverter. Considering all affecting factors such as material effi-ciency, integration, and structural configuration, boosting theMPPT capability is the most economical way of enhancing the ef-ficiency of the overall PV system [4].

Numerous studies have been undertaken to track the max-imum power point (MPP) from the output of PV systems subject touniform irradiance levels. The perturbation and observation (P&O)[5–9], hill climbing [10–14], incremental conductance [15–21],short-circuit current, open-circuit voltage, and ripple correlationmethods are the most popular of conventional approaches to MPPtracking. The main advantages of these methods are the use of asimple structure and fast convergence towards the MPP. Thesemethods however are only able to provide a reliable duty cyclesignal when a single MPP exists at the output of the PV system.Using bypass diodes within the circuitry of most of the current PVmodules increases the possibility of partial shading conditions(PSCs). The main consequence of these conditions is the occur-rence of multiple peaks at the output power–voltage locus. When

multiple peaks appear at the output because of PSCs, conventionalmethods fail to distinguish global MPPs (GMPPs) from local MPPs.The main reason for this failure is that the aforementioned tech-niques are based on hill-climbing theory, where the operatingpoint being followed shifts in the direction where the outputpower is maximized [22].

Considering these problems with partial shading, several stu-dies have been performed to modify the performance of conven-tional methods [23,24]. These modifications however are achievedthrough extensive mathematical computation, which can require apowerful and expensive controller. In addition, the modifiedmethods can only track the MPP under a limited number of PSCs.Given the unpredictability of environmental conditions, thesemethods can be unreliable in tracking the MPP. Different studieshave been recently performed where soft computing and artificialintelligence (AI)-based methods were applied to address the det-rimental effects of PSCs. The robustness, flexibility, and reliabilityof soft computing and AI make these methods highly suitable forPSCs. According to the growing number of studies published inthis field, a thorough study of recent developments in MPPTmethods is essential.

A number of valuable studies have been conducted to reviewthe performance of MPPT methods [4,25–34]. Most of these re-views however are limited to discussing the advantages anddrawbacks of the approaches without adequate detail to under-stand the procedures and structures of these methods. This lack ofdetail in the description is mainly due to the fact that the papersaim to cover all conventional and new approaches. Due to theemergence of PSC problems however, the selection of MPPTtechniques requires new considerations so as to make sure theyare applicable for environmental conditions typical to partialshading. Thus, most studies use AI-based MPPT methods and va-lidate these methods under different environmental conditions.This paper first focuses on the performance evaluation of recentresearch relating to six commonly used AI-based techniques cap-able of MMPT subject to PSCs. To complement the review of thesesix common techniques, this paper also discusses emerging tech-niques which despite only having been discussed in a limitednumber of publications, have demonstrated very good perfor-mance in addressing the shortcomings of other MPPT methodsunder partial shading conditions. The procedures and structures ofall the methods are described in sufficient detail to enable

M. Seyedmahmoudian et al. / Renewable and Sustainable Energy Reviews 64 (2016) 435–455 437

researchers and engineers to select the most appropriate MPPTtechnique.

Fig. 2. Output characteristics of the PV module under normal conditions: (a) I–Vcharacteristics and (b) P–V characteristics [35].

Bias Voltage

Cur

rent

Voltage

Unshaded cell

Shaded cell

Operation of unshaded cell in reverse bias

region

Break Down Voltage

Operation of unshaded cell

Fig. 3. I–V characteristics of PV cells in a reverse bias region.

2. Photovoltaic system characteristics

2.1. Normal conditions

Fig. 1 presents a single-diode circuitry for a PV cell. The outputof PV systems is directly affected by solar irradiance and tem-perature. Thus, the latest values of these factors should be em-ployed to obtain the MPP. In addition, the mathematical model ofPV changes with the short-circuit current (Isc) and open-circuitvoltage (Voc) were obtained from the data sheet provided by thecell manufacturer.

Hence, the generated power of a single solar cell is inadequatefor any convenient application. Thus, the cells in a PV systemshould be connected either in series or in parallel to enhance theoverall capability of the system, which ensures that all cells in thePV module (Ns, considering the given number of cells) contributeto the output power. The module output can be obtained throughthe following:

( )

= − −( + )

− −( + )

= ( )

⎡⎣⎢

⎛⎝⎜

⎞⎠⎟

⎤⎦⎥

F I V T G

I I Iq V I R

N AKTV I R N

R N

, , ,

exp 1

0 1

pv pv k

ph pv opv pv s

s k

pv pv s s

p s1

where IPV and Vpv are the output current and output voltage, re-spectively, to be added through the load or network grid; Iph is thesolar-generated current; Io1 is the diode saturation; A is the diodeideality factor; q is the electron charge constant (q¼1.602�10�19 C); and K is the Boltzmann constant. In Eq. (1) theparallel resistance (Rp) generally has a high value and sometimesassumes infinity in the PV module modeling because of the slightimpression of Rp. By contrast, the series resistance (RS) valuecannot be disregarded because of the effect of RS on output power.Fig. 2 depicts the output of the KC85T PV module in this study. Thismodule provides a power of 87 W at a terminal voltage and cur-rent of 17.4 V and 5.02 A, respectively.

2.2. Partial shading conditions

The entire PV system or some parts of this system may beshadowed by clouds, tall trees, and high buildings in outdoor en-vironments. This scenario causes non-uniform insolation condi-tions. The part of the PV modules that receive uniform irradiancecontinues operating at optimum efficiency during PSCs. Fig. 3shows that shaded cells are imposed to operate with reverse biasvoltage to provide the same current as the unshaded cells becausethe current that flows through every module in a series

+

-

Iph Rp

Rs

Io

Ipv

Vpv

Fig. 1. Equivalent circuit of a PV cell.

configuration is constant. However, the resulting reverse powerpolarity consumes power and reduces the extractable outputpower of the shaded PV system. Continuous operation of theshaded cells in excessive reverse bias voltage can result in hotspots and cause an open circuit in the entire PV array. This phe-nomenon is usually solved by inserting a bypass diode into apredefined number of cells in the series circuit [35].

Fig. 4 shows k-series-connected PV modules with the allocatedbypass diodes of the modules in the array. The characteristics ofthe PV system with the bypass diodes are different from thosewithout such diodes. Given that bypass diodes provide an alter-nate current path, the cells of the module do not have the samecurrent in PSCs. Therefore, the P–V curve creates multiple maxima.This occurrence is a crucial issue, and most common MPPT tech-niques cannot determine the difference between local and globalmaxima. Therefore, a proper MPPT technique should be able todistinguish the global maximum among multiple local maxima.Fig. 5 shows how extractable MPP differs in the shaded PV arraywith and without bypass diodes. In the following step, the

Fig. 4. Circuit model of an array that consists of k-series connected array.

Out

put P

ower

Voltage

MPP in PV with Bypass Diode

MPP in PV without Bypass Diode

Fig. 5. Power–voltage curve of a PV array under PSCs.


performance of common AI-based MPPT methods under normaland PSC conditions is evaluated.

3. Common SC based MPPT techniques

3.1. Artificial neural network

3.1.1. TheoryArtificial neural network (ANN) is one of the most reputed

methods among all soft computing methods that model the op-erations of biological neural systems. Basically, neural networksare collections of interconnected processing units called neurons,through which signals and information pass. ANN can be con-sidered a mathematical model of a brain-like system that func-tions as a parallel processing network. This system should undergoan extensive and careful training process to learn how to reflectthe input patterns in the actual system. Notably, any ANN system

is normally trained to function in predetermined systems andconditions because ANN neurons are trained to process and re-spond based on the in–out pattern relationship of the system.Therefore, for any system change, the designed ANN should un-dergo a retraining process to avoid producing unreliable and in-accurate responses [4,37].

An ANN does not require complex mathematical functions andphysical models when managing a system. An ANN can createpredictive functions from multidimensional data resources.Nowadays, ANN is used in a wide range of engineering applica-tions, such as pattern identification, environmental factor fore-casting, energy production estimation, and electrical load predic-tion. An ANN system is favorable for complex and ill-definedproblems, where fuzzy or incomplete information are involvedand a decision is normally made based on intuition. However, sucha system normally cannot handle problems that require high ac-curacy and precision [38].

The basic structure of an ANN comprises the following threelayers: input, hidden, and output. The schematic diagram of amultilayer feed forward structure is presented in Fig. 6. As shownin this figure, the neurons in each layer are connected throughsynaptic weights of the other neurons in the previous layers.Knowledge is usually stored as a set of connection weights (whichpresumably correspond to synapse efficacy in biological neuralsystems). Training is the process of modifying the connectionweights methodically by using a suitable learning method. Anetwork uses a learning mode, where an input is presented to thenetwork along with the desired output, and the weights are ad-justed for the network to attempt to produce the desired output.After training, the weights contain meaningful information,whereas the weights are random and meaningless before training[38].

3.1.2. ANN based MPPTIn ANN-based MPPT, the input variables are normally selected

as PV system parameters, such as short-circuit current, open-cir-cuit voltage, terminal voltage, output current, or environmentalfactors, which include irradiance level, module temperature, andwind speed [39,40]. These variables are received in the input

Fig. 6. Schematic diagram of the multilayer feed forward structure.


layers, processed in the hidden layers, and provide the requiredsignals in the output layer. Selecting the input variables and thenumber of nodes significantly depend on system complexity, de-signer knowledge, and data availability.

The capability and accuracy of ANN-based MPPT largely de-pends on the functionality of the algorithm in the hidden layersand training process. Several months or even years are required fortraining procedures to boost the functionality of ANN-based MPPTunder various environmental conditions. To monitor MPP withhigh accuracy and precision, the associated weights of the neuronsshould be obtained through an intensive and comprehensivetraining process. During this process, neurons are weighted inaccordance to the input–output relation of the targeted PV system.Therefore, the trained ANN for this PV system, which has a par-ticular specification, might not respond properly in other types ofPV systems. The output variable is normally the signals by whichthe operation point approaches the MPP region. The most com-mon output signal selected in literature to provide control signalsfor a DC–DC controller is the duty cycle or voltage Ref. [4].

The performance of the ANN method in identifying solutionsaccording to unknown parameters motivates researchers to em-ploy this method when monitoring the MPP while consideringenvironmental factors. Hiyama et al. [41] used the ANN approachin the MPP to track the system under normal conditions. In theproposed method of the authors, open circuit voltage is consideredthe input, while the required operating voltage serves as theoutput signal. Authors in [42] used a radial basis function networkto predict the MPP of a PV system. According to the authors, incontrast to conventional MPPT methods, the proposed method cantrack MPP without searching around the optimal power point. In[43], the authors propose the ANN peak power tracking of a PVthat supplies a DC motor that drives two different load torques. Inthis study, gradient descent algorithm is used to train the pro-posed ANN-based MPPT controller. Bahgat et al. proposed an ANN-based MPP tracking approach, as presented in [44], which cantransmit around 97% of the actual maximum generated power ofthe PV system under a uniform insulation level. In [45–48], theANN method is used to track the MPP using back-propagationneural network. In such systems, more nodes are used in thehidden layer, thereby possibly producing more accurate results.However, a controller cannot respond properly to fast irradiancechange conditions (FICCs) when using numerous nodes [4,49].

The application of ANN-based MPPT under FICCs has beenpresented in some research recently [40,50]. In [40], Liu et al.developed and proposed neural network to obtain the parameters

of an emulated MPP locus. According to the authors, the mainadvantages of the developed method are fast tracking speed andhigh dynamic and static efficiency to track the MPP under FICCs.However, the proposed method cannot work under PSCs. Laudaniet al. [50] proposed an MPPT algorithm based on neural networksembedded in a low-cost microcontroller. The output results con-firm the accurate performance of the proposed method underabrupt changes in environmental parameters. However, the vali-dation results are limited to the simulation investigations. Somereferences used ANN is used in order to improve the performanceof P&O method by detecting the variation in environmental con-ditions [51–53]. However, the proposed methods cannot monitorthe MPP if the PV system operates under PSCs. This drawback ismainly caused by the uncertain behavior of the PV system underPSCs and the dependency of the ANN training process on previousdata.

Given this problem, various researchers utilized the advanta-geous characteristics of the ANN method to optimize or tune theparameters of other established MPPT methods. For instance, Pu-nitha et al. [54] presented an ANN-based modified incrementalconductance method as the main approach to monitor the MPP ofa PV system under PSCs. This proposed method is validated bysimulating a field-programmable gate array-based experimentalverifications. The performance of the algorithm is compared withthat of P&O and fuzzy-based modified hill climbing algorithms.The results show the effectivity of the proposed approach intracking the MPP under PSCs. However, the accuracy and precisionof the output signal is still dependent on the ANN training process.Furthermore, the back-propagation learning rule is used to adjustthe network weights. Thus, the system may not function properlyunder abrupt climatological changes. Xu et al. [55] employed atwo-stage algorithm, where ANN uses the results of the traditionalincremental conductance method as the training data. The systemperforms satisfactorily under normal conditions and FICCs. How-ever, the system has not yet been tested under PSCs. In addition,the validation is limited to the simulation results. Ramaprabhaet al. [56] developed a method based on genetic algorithm (GA)-optimized ANN to enable a controller to operate under PSCs. Thismethod is validated under several simulated PSCs. However, thesystem cost increases when insulation and temperature data areused for the training process. A brief summary of studies em-ployed ANN method for MPPT unit is presented in Table 1.

Table 1The brief summary of some of the applied ANN-MPPT in the literature.

Method References Remarks

ANN [41–44] The standard ANN method is employed for the MPPT of the PV system under normal conditions, but the tracking fails under FICC.[45–48] Back-propagation neural network is used to increase the accuracy of the system.[40,50] The modified ANN is employed to track the MPP under FICC.[51–53] ANN method is employed in order to improve P&O functionality.[54] The ANN method is proposed to modify the incremental conductance method to track MPP under PSCs.[55] A two-stage algorithm is employed. Satisfactory results are obtained under normal conditions and FICC, but the verification fails under PSCs.[56] An ANN method is developed to optimize the GA technique to track the MPP under PSCs.

NB NS Z PS PB

Fig. 8. Membership function of FLC.

Table 2Rule base table of FLC.

ΔE E

NB NS Z PS PB

NB ZE ZE NB NB NBNS ZE ZE NS NS NSZ NS ZE ZE ZE PSPS PS PS PD ZE ZEPB PB PB PB ZE ZE


3.2. Fuzzy logic control

3.2.1. TheoryFuzzy logic control is one of the most commonly used techni-

ques in different engineering challenges because of its multi-rule-based characteristics [57]. Fuzzy logic control has a simple andclear procedure because exact mathematical modeling and tech-nical quantities of a system are not required for this controller[58]. In this technique, the variables manage some non-numericand linguistic actual values that range between completely trueand completely false values, such as high, low, medium, and often[59]. In fuzzy logic controller optimization, everything is definedaccording to the matter of degree, exact reasoning is replaced byapproximate reasoning, the mathematical model of this system isnot required, and decision is based on estimated values [37,60].

The technique comprises three different steps: fuzzification, ruleinterference diagram, and defuzzification. A block diagram of thistechnique is shown in Fig. 7. The input variables of a fuzzy logiccontroller highly depend on the controller design, but these vari-ables are generally known as error “e” and variation in error “Δe.”The definition of these input variables, given the fuzzification stepsand the output signals received from the defuzzification step, isdetermined by the requirements of the targeted problem. The threemain stages in the fuzzy logic process are described as follows:

3.2.1.1. Fuzzification. The input values in this stage are fuzzified bysome predefined membership functions. The main task of fuzzi-fication is to convert crisp values to linguistic values via mem-bership functions. Different membership functions exist, such astrapezoidal, Gaussian, and triangular. However, the triangularmembership is the most popular, as shown in Fig. 8. The acronyms“NB,” “NS,” “Z,” “PS,” and “PB” in this figure represent negative big,negative small, zero error, positive small, and positive big, re-spectively [61,62]. In general, the quantity of membership func-tions is more effective than the type and shape in terms of systemspeed and accuracy. More membership functions, which are re-commended for more complex engineering problems, result inhigher accuracy and longer processing time. By contrast, fewerdesigned membership functions result in faster processing timeand higher possibility of diversity.

Fig. 7. FLC bloc

3.2.1.2. Rule interference. This stage is designed to control theoutput variables according to an inference engine. The inferenceengine applies rules to membership functions using a rule basetable. Five rule-based inferences are shown in Table 2. The designof these rules is based on the “if–then” concept and requires de-signer knowledge about the systems. Mamdani's inference methodis the most common method that uses the max–min operationapproach.

3.2.1.3. Defuzzification. Membership functions are once again em-ployed to convert output linguistic values into relevant crisp andnumerical values. However, the range of the membership func-tions used in this section may vary based on designer knowledge.Several defuzzification techniques exist, such as center of area andmean of maxima; the latter is more common in control

k diagram.


applications [63].

3.2.2. FLC-based MPPTIn MPPT applications, the input variables of the fuzzification

step in an FLC system can be calculated by using the followingequations:

( ) = Δ ( )Δ ( )

= ( ) − ( − )( ) − ( − ) ( )

e tP tV t

P t P tV t V t

11 2

Δ ( ) = ( ) − ( − ) ( )e t e t e t 1 3

where t is the sampling time; P(t) and V(t) are the power andvoltage values at the operating point, respectively; Δe(t) definesthe direction of the operating point in the next movement of theoperation; and e(t) is the position of the operation point in the P_Vlocus with respect to the maximum power point.

In the next section, the fuzzy rules, as mentioned in Table 2, areused to determine the MPP based on “THEN Changes applied ANDpower increased THEN continue the direction” to define the requiredchange in duty cycle of a DC–DC converter. Finally, defuzzificationconverts the output of the rule base section into non-fuzzy values.In MPPT applications, only the centroid method is used to definethe proper duty cycle variation as the output of the MPPT con-troller based on fuzzy logic scheme.

3.2.2.1. Early research. According to literature, different re-searchers have tested the fuzzy logic-based MPPT method in var-ious PV systems and under different environmental conditions[64,65]. In early studies, the authors in [66] introduced the fuzzylogic controller as MPPT, where dPpv/dIpv and its change [Δ (dPpv/dIpv)] are considered fuzzy controller inputs. The main problem ofthe proposed method is the inefficiency of the dynamic responseunder FICCs. Simon et al. [67] improved the dynamic response offuzzy controller through duty cycle and power variations (ΔDk andΔPpv) as inputs for the system. However, the accuracy of the sys-tem is still limited under varying irradiance and temperatureconditions. Methods of improving fuzzy logic controller for betterdynamic characteristics and accuracy have been introduced by theauthors in [63] and [68]. In [63], the authors selected three inputvariables, namely, dPpv/dIpv, [Δ (dPpv/dIpv)], and duty cycle variation(ΔDk), whereas temperature and irradiance values are the inputsin [68]. Both systems show an accurate dynamic response underenvironmental parameter variations. In [69], the authors con-sidered the instantaneous values of voltage and power at theoutput of the PV system as the input of the FLC system and verifiedthe efficiency of the proposed method for a stand-alone PV waterpumping system. Liu et al. [70] designed an FLC-based MPPTmethod according to an asymmetrical membership function con-cept. The final results show improvements of 42.8% and 0.06% intransient time and higher MPPT tracking accuracy, respectively,compared with the symmetrical FLC-based MPPT approach.However, similar to previous references, the proposed methodshave the following significant drawbacks [71]: first, these methodslargely depend on the shape of membership functions and the rule

Table 3The brief summary of some of the applied FLC-MPPT in the literature.


FLC [63,66–71] Standard FLC was for MPPT application. These method were smembership functions. In addition the methods were unable

[72–79,65] Optimized FLC was employed. The methods were more inde[64,83,54,59] Hybrid methods including FLC and other methods such as P&

is improved. The complexity increased.

base interference. Second, the designers of these methods requirea comprehensive knowledge of the PV system operation; in-sufficient knowledge would usually cause an inefficient and slowsystem.

3.2.2.2. Reduction of designer dependency via the optimized FLC. Toreduce the effects of the aforementioned drawbacks, some re-searchers have attempted to optimize or combine normal fuzzylogic controller with other methods. In [72] and [73], the authorsoptimized the performance of fuzzy logic controller through GAand particle swarm optimization (PSO) approaches, respectively. In[65], the authors improved the performance of normal fuzzy logiccontroller together with the fuzzy cognitive network. In such asystem, different nodes exist, which represent the operational andcontrol variable of a PV system. Node interconnection weights aredetermined by using data to cover the PV system under differentenvironmental conditions. When the system is trained, the systemcan be mounted in any PV system. The method has a fast trackingspeed under normal conditions. In [74–79] the FLC in combinationwith other methods has been used in order to increase the systemindependency of MPPT unit. In [80] authors used a single inputfuzzy logic controller to reduce the complexity and hence achieveda faster and easier implementation. Authors in [81] used anadaptive neuro-fuzzy solar cell model to estimate the function ofthe array junction current in order to improve the efficiency ofarray power and reduce the hardware setup. However, likemethods presented in [65,72–80], it cannot, or at least have notbeen verified, to track the MPP under PSCs.

3.2.2.3. Performance under partial shading conditions. Recently,several researchers have used the FLC approach to improve theperformance of MPPT controller under PSCs. In [82], fuzzy logic isapplied to define the perturbation size in a proposed P&O tech-nique. In [64], fuzzy logic controller is used to enhance the per-formance of the hill climbing method by scanning and storing theMPP during the P&O procedures. The proposed methods in bothreferences perform satisfactorily in determining the GMPP undercertain PSCs. The authors in [83] optimized normal fuzzy logiccontroller by using the GA approach, which has been successfullytested under PSCs. In [59], FLC with ANN is employed to track theGMPP. In this method, irradiance level and cell temperatures arethe main inputs to train the ANN process to determine the MPP.However, this information cannot be acquired in some shadingconditions. However, these methods require extensive computa-tions, which make them too complex for commercial use. In ad-dition, this method requires extensive computations in the fuzzi-fication, rule base, and defuzzification stages. The highlights ofstudies employed FLC-MPPT have been presented in Table 3.

3.3. Particle swarm optimization

3.3.1. TheoryPSO is one of the metaheuristic search tools that receive con-

siderable attention in engineering applications. PSO was first in-troduced in 1995 by Kennedy and Ebrahat; this method is inspiredby the natural behavior of birds flocking [84–86]. This theory

ignificantly dependent on the designer knowledge about PV system and also theto track MPP under PSCs.

pendent and very fast, however unable to operate under PSCs.O, ANN and GA were designed and tested. The ability to operate under some PSCs

0 20 40 60 80 100 120 140 160 180 2000

0.2

0.4

0.6

0.8

1

1.2

1.4

Iteration

Iner

tia W

eigh

t

Fig. 9. Behavior of inertia weight during all the iterations.


basically explores a specific area called the solution space, whereeach location has a possibility degree for the solution of problems.The PSO moves each particle throughout the solution space todetermine the optimum solution according to the individual andneighboring particle experiences of the PSO during optimization.Therefore, particles involved in optimization use the memory ofthe particles to modify the particle fitness by following the be-havior of the successful particles in the swarm.

The PSO procedure starts with a random particle (initializa-tion), continues by searching for optimal solutions within the pastiterations (movement), and then evaluating the particle qualityaccording to the fitness function (evaluation) [87]. The main stepsof this algorithm are described as follows.

3.3.1.1. Initialization. In this step, PSO defines the population sizeby randomly selecting the participant particles during optimiza-tion. The particle are selected from D-dimensional solution space,where D is determined based on the number of variables for op-timization. Given the unavailability of information about the sys-tem operation, the particles in the initialization step are randomlyselected and optimized through the succeeding steps of thealgorithm.

3.3.1.2. Movement. The current position (Xi) of each particle is leftwith the speed of (Vi) to stochastically explore the search spaceand determine a better solution. The movements of the particlesare in accordance with the best (Pbi) position and global best (Gb)position of the particles. Particularly, Pbi is the best position ex-perienced by the i-th particle throughout all previous iterations,while Gb is the best position experienced by the sum of all theparticles within all the past iterations.

During optimization, the particles adopt the values of the ob-jective function, while the Gb and Pbi of the particles are recorded.The basic PSO algorithm that defines the next position of thecandidate solution is as follows:

( ) ( )= × + × × − + × × − ( )+V w V r c P X r c G X 4ik

ik

bi ik

b ik1

1 1 2 2

= + ( )+X X V , 5ik

ik

ik1

where i represents the optimization vector variable, k is thenumber of iterations, Vik and Xi

k are the respective velocity andposition of the ith variable within k iterations, w is the inertiaweight factor, c1 is the cognitive coefficient of the individual par-ticles, c2 is the social coefficient of all the particles, and r1 and r2are the random selected variables in the range [0, 1]. These ran-dom parameters mainly aim to maintain stochastic movementwithin the iterations. To maintain the search space in a certainarea, the velocity values are set to the range of [0, Vmax].

In PSO, the parameters w, c1, and c2 are highly mutable. A slightchange in the values of these parameters may affect the speed andaccuracy of the algorithm. The optimization may involve the localmaximum through any poor design for c2, and the optimizationaccuracy can be diminished by the inappropriate value of c1. Si-milarly, the influence of inertia weight on the speed and con-vergence of PSO is substantial; a large value for this parametercauses slower convergence, whereas a small value produces anarrower search space range. In conclusion, a shift in the values ofinertia weight encourages the particles at the initial stages todiffuse before gradually limiting the search space in the finaliterations. The behavior of the inertia weight values is shown inFig. 9.

3.3.1.3. Evaluation. The fitness of the particles in the new locationis evaluated and saved to improve the particle movements in the

succeeding iterations. The best self-experienced location (Pbi),which records the best position experienced by the ith particle upto the current iteration, is updated once Eq. (4) is satisfied. Inaddition, the variable (Gb) records the Gb experienced locationmet by all contributed particles throughout the past iterations andis compared with the Pbi in each iteration. The update condition ofGb and Pbi is presented in Eq. (7) and Eq. (8). These equationsindicate that Gb is only recorded as Pbi if the conditions below aresatisfied.

( ) ( )= ≥ ( )P X If F X F P , 6bi ik

ik

i

( ) ( )= ≥ ( )G P If F P F G . 7b bi bi b

The particles continue to operate until the stopping conditionsare met. This condition is designed based on the complexity of thesystem, required accuracy, and time limitation of the controlprocess. The flowchart of described steps of PSO is presented inFig. 10.

3.3.2. PSO-based MPPTIn the MPPT application, the search space depends on the

configuration of the PV system. For example, if an MPPT controlleris designed to control D number of converters, the search spacewould be D-dimensional, where each dimension represents avoltage value as a solution to the problem. The evaluation of theparticles is based on the output power of the PV panel, which isdenoted by F as the fitness evaluator. Eq. (8) shows the locationmatrix of the particles as a solution to the MPPT problem in the D-dimensional search space.

( )= ( )−⎡⎣⎢

⎤⎦⎥X X , X , X , ..... , X , ...... X , X , 8i

kN1

k2k

3k

ik

1k

Nk

where Xik is the location of the ith particle at the kth iteration.

Practically, the generated power fluctuates because of variations inthe insolation levels and partial shading degrees. Therefore, thealgorithm should be initialized when Eq. (9) is satisfied. If this stepis neglected, the actual values of the acquired Pbi and Gb valuesshould not be considered.

( ) − ( )( )

> Δ( )

+F X F XF X

P9

i i

i

1

3.3.3. Discussion and previous studiesThe idea of using PSO approach as a tool for tracking the MPP

initially emerged to improve the efficiency of MPPT controllerunder PSCs. Tracking the global MPP by using the PSO methodmakes the controller system independent, fast, and robust in PSCs.The PSO-based MPPT approach has been presented in severalstudies [22,26,88–90].

Fig. 10. Flowchart of PSO.


3.3.3.1. Early studies. A highly efficient PSO-based MPPT was em-ployed by adding repulsive force to the PSO algorithm to track theMPP under climatological changes [88]. The study proves the re-liability of the proposed technique under frequent environmentalchanges. The PSO approach has been proposed to track the globalmaximum from the output characteristic of a modular PV system[22,89]. The system cost has been significantly reduced as thenumber of sensors is reduced in the experimental implementa-tions. However, using standard PSO imposes several limitations onthe control system. In the standard PSO, high velocity must beused for a particle whose position is too far from the position ofthe best particles (Gbest), which is defined as the best experiencedposition of all particles. In addition, the trajectory of the particles islimited by designed acceleration. Low acceleration results in asmooth trajectory for particles but slow convergence. By contrast,high acceleration may accelerate computation but may lead theparticles toward infinity [91]. In addition, a random coefficient inthe cognitive and social parameters of the PSO component sig-nificantly affected the performance of the algorithm, which re-quires intensive knowledge for optimum parameter tuning. Thelow-valued random coefficients result in slight movement of theparticles. Therefore, more iterations are required to find the opti-mum values. On the other hand, large-valued random coefficientsincrease the step size of particle movement, which can lead themout of the search space margin or in the local MPP. In addition, theinitial location of the particles significantly affects the convergencequality and speed of the algorithm.

3.3.3.2. Improved version. Some studies have modified the stan-dard PSO technique [92–94]. Chowdhury et al. [92] modified theperception radius and search direction of each agent in accordance

Table 4The brief summary of some of the applied PSO-MPPT in the literature.


PSO [22,88–90] Standard PSO was employed in the different PV syste[92] Convergence and accuracy was improved by modifyin[93] Steady-state oscillation was reduced by using the PSO[94] DPSO was used. Convergence speed and accuracy wa[95] Combination of P&O and PSO was used. Convergence

with its performance improvement throughout algorithm itera-tions. In the adaptive perceptive PSO method, all agents shouldscan their own range of the search space. The modification sig-nificantly increases the accuracy of the algorithm to find the globalmaximum. However, in this method, the additional dimensionalsearch space leads to an assignment of a higher number of parti-cles and consequently boosts the computational burden andcomplexity of implementation. Ishaque comprehensively studiedthe improvement of the PSO technique in [93] and [94]. In [66],steady-state oscillation was reduced by using the PSO method inconjunction with the direct duty cycle method. In [94], a fast andaccurate global tracking method was achieved by using a proposeddeterministic PSO method. In this method, the convergence andthe accuracy of the PSO-based method to find the global MPP wasenhanced by eliminating random numbers. However, the removalmight reduce the reliability of the method under intensive mis-matching condition as the algorithm loses the main advantage ofevolutionary algorithms and cannot track the global maximumpoint in all partial shading patterns. In [95], a P&O method wasemployed to improve the convergence speed of PSO. In thismethod, the first P&O identifies the nearest local MPP and then thePSO starts searching to find the global MPP. As a result of using thiscombination, the search space exploration is reduced in earlyiterations. In [36] hybrid method called DEPSO, a combination ofPSO and DE, is used in order to reduce the impacts of randomcoefficients and enhance the system independency of standardPSO method. The proposed technique showed a satisfactory per-formance under PSCs. However, the problem of this method is theoscillation during tracking period. Table 4 presents a brief sum-mary of studies applied PSO-MPPT in the literature.

m configurations and under various partial shading patterns.g the search direction of agents in a proposed adaptive perceptive PSO method.in conjunction with the direct duty cycle method.

s improved.speed was improved.


3.4. Ant colony optimization

3.4.1. TheoryAnt colony optimization (ACO) is a probabilistic algorithm used

for global solution search in a stochastic problem. The algorithmwas first introduced by Dorigo and Gambardella in 1997 based onthe foraging behavior of actual ants to find the best path towardfood [96]. Shen et al. [97] and Dorigo et al. [98] modified the al-gorithm and presented it in the form of an optimization method.Basically, ACO mimics the social behavior of ants searching for afood source. First, all individuals search for the food. Once any ofthe ants finds food, it estimates the size of the food. If the food istoo large, it takes a small portion and brings it to the nest. The antleaves pheromones in its path as it moves toward the nest. Thesepheromones constitute a pheromone trail, which is used by otherants to find the source. The density of the pheromones in the pathis directly related to the number of the ants that travel the path. Asmore ants pass along the mentioned path, more pheromones aredeposited [99]. Given that the pheromones are vaporized overtime, the density of the pheromones would be reduced if otherants do not travel the same path. Therefore, the shorter path ismore preferred by ants. This methodology leads the ants to findthe shortest possible route from the nest toward the food sourceby exchanging such information among team members. In general,the ACO trail will be selected and updated if it represents the goodproblem solution [100]. Fig. 11 illustrates the behavior of antswhen finding food. The figure is divided into 6 sections and showsthe procedure of selecting the best trail between nests and foodsources.

Two ant colony schemes have been defined in the literature.The application of ACO for discrete optimization problem has beendiscussed in [102,103] and the continuous problem in [104,105]. Incontinuous ACO, the pheromone density (τt) is intensified aroundthe best objective that is found during the algorithm operation.This density then leads the ants to converge toward the bestpossible path obtained in the solution space. Eq. (10) defines thelocation of the ant in the search space.

= ± ∂ ( = ) ( )+x x x where t T12, 3, ... 10tk gbest

1

in which T represents the predefined number of iteration, t re-presents the current iteration number, Xgbest represents the bestobjective location value that has been found by the algorithm upto the current iteration, and ∂x is the randomly elected vector inthe area of [-α, α], which denotes the variation allowance of the antfrom the best objective value (Xgbest). Following the studies in[106,107], the length and direction of variation in accordance tothe best objective value can be defined by Eqs. (11) and (12), re-spectively, after termination of T iteration. The direction of theant is

Fig. 11. Ants behavior

α α= × ( )+ 0.1 11t t1

( )¯ = + ( × ) →

(¯ ) − ( ) ≤ → ( + )

( ¯ ) − ( ) ≤ → ( − )

⎪

⎪

⎧⎨⎩ 12

x x xif f x f x sign

if f x f x sign0.01

0

0best gbest gbest

best gbest

best gbest

The deposited pheromone then vaporizes. The simulation ofthe vaporization is shown by Eq. (13), while the pheromone in-crement around the best objective is presented in Eq. (14).

τ τ= × ( )−0.1 13t t 1

τ τ= + ( × ( )) ( )− f x0.01 14t tgbest

1

A brief and simplified description of ACO is presented step bystep in Fig. 12.

3.4.2. ACO-based MPPT techniquesIn the ACO-based MPPT technique, each location represents the

voltage value of PV array. In addition, f(x) objective function is theoutput power of PV array by which the fitness of each ant isevaluated according to its respective voltage value. Assuming thatN ants participate in the optimization, the following Eq. (15) showsthe location matrix for ants in tth iteration.

= … ( )⎡⎣ ⎤⎦X X X X, , , 15t

it t t

N1 2

The variable environmental conditions and changes in shadingpatterns must be considered by using the optimization procedure.Consequently, the condition presented in Eq. (16) results in re-initialization of the algorithm once it has been satisfied.

( ) − ( )( )

> Δ( )

+F X F XF X

P16

i i

i

1

ACO has recently been used to solve problems in different re-newable energy sectors [101,104]. In [108], ACO is applied to findthe global MPP from the output of a modular PV system. Themethod was tested under four predefined shading conditions andshowed satisfactory results under these conditions. The proposedmethod was fast and has a low system cost. It has the ability totrack global MPP by using a single current sensor at the output ofthe PV system. In addition, unlike PSO, the convergence speed ofACO is independent of the initial conditions. However, the ver-ification part was limited to simulations, and the experimentalfeasibility of the system under real conditions was not demon-strated. In addition to the sole application of ACO for the MPPTproblem, some researchers used the advantages of the ACO ap-proach to design hybridized methods in which ACO is employed tooptimize the performance of other methods. For instance, in [109],ACO is employed to optimize the parameters of an adaptive FLC-based MPPT controller to improve the dynamic response of the

in six steps [101].

Fig. 12. Flowchart of ACO method.

Table 5The brief summary of some of the applied ACO-MPPT in the literature.

Method Reference Remarks

ACO [107] ACO method was applied for a modular PV system. Its capability under PS was verified by using a single sensor.[108] ACO was employed to optimize the FLC parameters. Steady-state errors and dynamic response were improved compared with normal FLC.

Its capability under PS conditions was verified. Its complexity was increased.[106,108,110,115] ACO was employed to FLC gains in the AAF_MPPT method. Its tracking capability under FICC is verified. The complexity was increased. Its

capability under PS was not verified. Experimental verifications were lacking.


system, as well as reduce steady-state errors. The satisfactorydynamic and steady-state performance of the simulated PV systemeven under variable irradiance levels reflects the superiority of theproposed control technique under any environmental condition. In[110] ACO was used to design the FLC gains in an ant adaptivefuzzy logic MPPT technique. The proposed method achieved agood response even under extreme irradiance variations. However,the simulated results were verified based only on the performanceof the system under uniform insolation levels. The highlights re-lated to the ACO-MPPT techniques, presented in the literature, arelisted in Table 5.

3.5. Genetic algorithm

3.5.1. TheoryGA is a meta-heuristic optimization method for finding solu-

tions based on evolution biological behavior. This method wasintroduced by Holland in 1975 through the principle of survival ofthe fittest [111]. In general, GA models the candidates’ solutions,known as chromosomes, in the problem search space throughfixed-length strings. A chromosome consists of a constant popu-lation of genes that can be presented either as real or binary codes.In the algorithm procedure, the population of chromosomesevolves gradually over generations through competition [37]. Thisevolvement is achieved through the following applications of ge-netic operators: selection, crossover, and mutation. The evolutionprocess helps the best fitness chromosome to survive and matefrom one generation to the next [37]. In other words, the GAO is aniterative method through which a population of chromosomes

evolve and improve through generations affected by GA oper-ators. In each generation, parents are selected from the existingpopulation and used to generate children in the next generation.The objective function is the main evaluative factor for improvingpopulation fitness over time. The procedure and life cycle of thepopulation are shown in Fig. 13, and the following steps are theimplementation procedure of GA optimization.

3.5.1.1. Initialization. First, the objective function is defined as themain tool for ranking the candidate solutions’ fitness throughoutthe algorithm procedure. The selection and formation of an ob-jective function is based on the requirements and complexity ofthe targeted system. Then, the first population of chromosomes isdefined through random selection. The convergence speed of thealgorithm largely depends on the population size. Similar to anymetaheuristic method, a large population is recommended for acomplex optimization problem; however, it dramatically reducesthe convergence speed of the algorithm.

3.5.1.2. Selection. Initially, the fitness values of the chromosomesin the initial population are evaluated by defining an objectivefunction. Then, based on the fitness evaluation results, the chro-mosomes are included in the current population. In general, thechromosomes with the higher fitness value will have an increasedchance to be considered for the next generation.

3.5.1.3. Crossover. Two selected chromosomes are combined in thisstep. The application depends on how the genes in the chromo-somes are coded. The normal crossover technique can be used if

Fig. 13. Life-cycle of population and recombination and mutation [37].


the genes are in binary format. The random integer number, whichis smaller than the number of genes in the chromosome, is used tosplit parent chromosomes and generate offspring, as shown inFig. 13 [37,112,113]. If the chromosomes are coded in continuousnumbers, and α is considered as crossover rate, the offspring willbe crossed over as below.

α α

α α

= × + ( − ) ×

= ( − ) × + × ( )

Offspring parent parent

Offspring parent parent

1

1 17

1 1 2

1 1 2

3.5.1.4. Mutation. The mutation operator maintains the stochasticnature of the algorithm by maintaining the genetic diversity ofcurrent generation to the next one. The mutation operator pro-vides chances for candidate solutions that cannot be obtained onlyby using the crossover operator application in the binary format.This operator randomly manipulates the chromosome with aspecific rate called mutation rate (β). For the continuous coded

START

DE Initialization

Generate the first population of the Choromosomes

Perform Cross over and generate

offsprings

Next iteration

Evaluate the populations’ fitness

Perform Selection to produce the parrents

GA

Mutate the chromosomes according to mutation rate

Fig. 14. Flowchart

chromosomes, the mutation follows the statement below [114]

β= ± + ( )Offspring Offspring Offspring. 18

The algorithm continues to work until the stoppage condition ismet. This condition can be set in regards to the requirement andthe complexity rate of the optimization problem. The brief pro-cedure of GA algorithm is shown in Fig. 14.

3.5.2. GA-based MPPTGA optimization has been used in different optimization pro-

blems faced by the renewable energy sector, such as MPPT. GA hasa rapid convergence speed in different optimization problems. Inaddition, the method showed satisfactory performance in reducingthe chance of being trapped in the local optimum. However, thismethod is not recommended for optimizing very large or ex-cessively complex problems. In the application of GA for MPPTsystem, the initial parent population is shown as:

All chromosomes finished?

Yes

Output the offspring

Re-initialization condition

is satisfied ?

reinitilization

No

No

Yes

END

of GA method.


= ( )…⎡⎣ ⎤⎦X Parent Parent Parent 19i N1, 2,

where n is the population size, and each parent represents initialvoltage values in which the algorithm starts the evaluation pro-cess. The objective function f (xi) is the generated power at theoutput of the PV system. The fitness values of each position areevaluated by the objective function and are used to evolve thepopulation and improve the population fitness through the gen-erations. In the MPPT application, the algorithm must be re-initialized because of sudden changes in load, solar irradiance, orpartial shading patterns. Therefore, the following conditions re-initialize the GA-based MPPT algorithm once they have been sa-tisfied:

( + ) − < Δ ( )V k V V1 20

( + ) − ( )( )

> Δ( )

P k P kP k

P1

21

3.5.2.1. Discussion and previous research. Among the mentionedartificial intelligence methods, GA has been used the least in theMPPT problem. Therefore, few references that used GA as the mainMPPT method are available. In one of the studies in [115], GA wasapplied to find the global maximum power point of the PV systemunder some simplified PSCs. In [116], an MPPT controller was de-veloped based on the GA approach, and the proposed method wasverified through two different case studies, each presenting dif-ferent PS patterns. However, the verification part of both studies islimited to simulation validations, and the feasibility of the methodunder an actual practical environment has not been presented. Inanother study [84], the performance of the GA-based MPPTmethod was compared with conventional P&O and the incre-mental conductance method under two predefined PSCs. How-ever, as in previous work, the verification part is limited to simu-lations. In addition, the mutation steps are eliminated in the pre-sent paper. Therefore, the stochastic characteristic of the designedGA is diminished. In [117], Daraban integrated the P&O methodwith GA, reduced the population size, and decreased the numberof iterations. The proposed method shows a faster convergence, aswell as a more accurate output, for a PV system under variouspartial shading patterns. It is compared with a normal GA ap-proach presented in [112].

The GA method has been mostly used in hybrid methods toimprove the performance of other MPPT techniques. For instance,in [118], GA was used to tune the parameters of a fuzzy logiccontroller used in MPP tracking under PSCs. The performance ofthe fuzzy logic controller improved as parameters, such as rulebase and membership functions, are tuned to their optimized va-lues by using GA techniques. In another study [119], the GAmethod was used as a tool to train the ANN system. In this ap-proach, the GA is used to define the neuron numbers in a multi-layer perceptron neural network. Table 6 presents a brief summaryof studies employed GA technique as a main strategy of MPPT unit.

Table 6The brief summary of some of the applied GA-MPPT in the literature.

Method Reference Remarks

GA [115,116] GA approach is applied to address the MPPT problem. It perform[112] GA performance is compared with the performance of conventi

experimental verification.[117] A modified GA method is integrated with a P&A approach. Its c[118] GA used to tune FLC parameters. Its performance is experiment[119] GA used to train the ANN parameters. Its performance under U

3.6. Differential evolution

3.6.1. TheoryAnother particle-based optimization method is the differential

evolutionary (DE) algorithm, which is used for global optimizationapplications. The algorithm has a similar concept to GA and wasfirst introduced by Storn and Price [120,121]. Owing to its sim-plicity, DE is one of the most powerful population-based optimi-zers. The problem is optimized by creating new candidate solu-tions based on different formulas for each of the different tech-niques while maintaining the population size. In this algorithm,existing particles with the best fitness records remain in the po-pulation, while the others are replaced by new particles. The al-gorithm is suitable for problems with many local optimum solu-tions [122–124]. The algorithm consists of the following foursections:

3.6.2. InitializationThe algorithm must be initialized by assigning the initial lo-

cations to the target vector. The parameters in the initializedvector are recommended to be selected randomly to enhance thestochastic nature of the method and cover the entire search space[125]. Assuming the NP as the total population size, Eq. (22) de-scribes Xi,G as the target vector of the ith particle in the Gthiteration.

∈ ( ) ( )x i NP; 1, 2, 3, .. .. 22i G,

3.6.3. MutationThe vi G, , which is the mutant or donor vector of the ith particle

in iteration G, is formed by mutating each particle of the targetvector by using the equation below

= + ⋅( − ) + ⋅( − ) ( )+v x K x x F x x 23i G i G r G i G r G r G, 1 , 1, , 2, 3,

where r1, r2, and r3 are the random numbers in the range of (1, 2,…NP), and nop denotes the number of particles. K and F are thescaling and combination factors, respectively, which are ranged in[0, 2].

The DE parameters, like other metaheuristic methods, affectthe performance of the system significantly. Parameter selection isan important responsibility for a designer, especially when theengineering problem is significantly critical, such as MPP trackingfrom the nonlinear output of partially shaded PV system. The mostcommon parameter selection method is rule of thumb by Storn[126].

3.6.4. CrossoverAfter generations of mutant vectors in the mutation section,

combining the target vectors and mutant vector generates the trialvectors through a non-uniform crossover operation as shown inEq. (24). The crossover enables the algorithm to propose bettersolutions by shuffling the data of successful combinations. In thisequation, ∈ { }j D1, 2, ... , , rndj is a random value between 0 and 1,and rni is a random number chosen from the domain of j. The

s well under some PS conditions. However, it lacks experimental verification.onal methods. It demonstrated tracking capability under PSCs. However, it lacks

apability under PSC is experimentally proven.ally verified under UIL. However, its capability under PSC is not verified.IL is experimentally verified. However, its capability under PSC is not.


maximum dimension of the search space is denoted by D. CR is theso-called crossover constant in the range of [0, 1], which is used inthe condition statement of the crossover section. If this value isgreater than the random number, the value of the trial value fol-lows the mutant value. If not, it follows the target value. In the DE-based MPPT, the output power will be evaluated when the cross-over sections are completed and the trial components of trialvectors are generated.

( )( )

=≤ =

> ≠ ( )+

+

⎪⎪⎧⎨⎩

uv if rnd CR or j rn

x if rnd CR or j rn 24ji G

i G j i

i G j i, 1

, 1

,

3.6.5. Selection

=( ) ≥ ( )( ) < ( ) ( )

+ ⎪

⎪⎧⎨⎩x

u if f u f x

x if f u f x 25i G

i G i G i G

i G i G i G, 1

, , ,

, , ,

In the final step, the selection operation defines the nextmovement of the particle toward the prospective region of thesearch space. The selection of the parent target vectors is in-dependent of the fitness values. However, the produced children(offspring) in the mutation and crossover must be evaluated by thefitness function and compared with the parents. The selectionoperator compares the fitness values of obtained trial vectors withtarget vectors. The parents remain in the population if they havebetter fitness values than their children. If not, the selection op-erator will replace them with a trial vector that has better fitness.

Fig. 15, shows the process of obtaining a new proposal in the DEalgorithm. Locations 1 and 2 in the search space denote twomembers whose difference vector is used to move a third popu-lation member (3). The obtained location, denoted by (4), is sub-jected to crossover with the replacement candidate, denoted by(5), to generate a new proposal (6). Afterward, if the proposal has abetter fitness value than the candidate, the candidate is updatedby the proposal. In the MPPT scheme, Ui,G represents the duty cyclevalue in the trial vectors; therefore, f (Ui,G) represents the outputpower at this duty cycle. The measured power for the duty cyclesin the trial vectors is compared with duty cycles in the targetvector. The result of this comparison will update the duty cycle atwhich the higher output power was achieved. Fig. 16, shows aflowchart of the process.

Fig. 15. Proposal derivation in DE [124].

3.7. DE-based MPPT

The DE method was recently applied in different extents tosolve the control issues of renewable energy systems, includingthe problem of global MPP tracing in PSCs. In the DE-based MPPtracking method, the target vector is normally considered the dutycycle of the designed DC-DC converter. The DE approach in theMPPT problem was first presented in [127,128] where a standardDE algorithm was used. However, the method is based on staticobjective function in which the P_V curve must be predetermined,which makes the method impractical for real-time MPPT appli-cation [122]. The DE method was used in MPP tracing of the PSCs[94,123]. The search time was reduced by avoiding the additionalmeasurement from the output of the converter setup. As men-tioned in the crossover step, the value of the trial vector compo-nents might be the same with the target value depending on thecrossover condition. Thus, the processing time was reduced byavoiding the measurement of the power for those duty cycles thatremained the same after the crossover step. Moreover, the pro-posed DE method was compared with the conventional HCmethod. The standard DE algorithm was modified to suit thenature of a dynamic tracking system. According to the author'sinvestigations, the proposed DE algorithm has the following sig-nificant features:

a) Simple and straightforward implementation.b) The ability to search the global maximum point regardless of

the initial parameter values.c) Fast convergence.d) Requires only a few control parameters.

The final results proved that unlike the conventional method,the proposed DE techniques have fast and reliable tracking per-formance under PSC without any oscillation at the output wave-form. However, the performance of the method was verified onlythrough simulations and experiments. Therefore, the responsetime and the complexity of an actual experimental setup have notbeen considered in the final verification of the method. In [129], atwo-stage MPPT technique based on the combination of DE andANN is proposed for a simulated PV system under PSCs. In the firststage, the DE method is used to find the area in which the MPPexisted. In the second stage, the ANN method is employed to trackthe exact MPP at the output Ipv-Ppv characteristics. The methodperformance was promising under different PSCs. However, theuse of a two-stage technique results in further complication in theoverall systems. In addition, as in previous studies, the proposedmethod was evaluated under simulated conditions only. High-lights and main points of studies, which used DE-MPPT in theliterature, are summarized in Table 7.

3.8. Emerging methods

This paper has thus far reviewed 6 popular AI-based MPPTtechniques including their ability to perform subject to partialshading conditions. Aside from these techniques, there are emer-ging approaches for MPPT subject to partial shading which haveonly very recently been proposed and discussed in a limitednumber of publications. Despite not having been extensively in-vestigated in the literature, these emerging approaches do presentstrong approaches to global MPP tracking, and are likely to be thefocus of future research attention. As such, in order to present acomprehensive review of AI-based methods for MPPT, theseemerging approaches are reviewed in this section.

3.8.1. Radial movement optimization (RMO)RMO is a swarm-based stochastic optimization technique [130]

START

DE Initialization

Min J Obtained? OR

Max iteration reached

Select the target vectors and perform

DE operation

Calculate fitness J

Yes

Next iteration

Output the best target vector

Mutation to obtain the mutant vector

Crossover to obtain the Trial vector

Re-initialization condition

is satisfied ?

reinitilization

Yes

No

No

NoYes

END

DE

Update the target vectors

Any improvement in\the fitness of trial vectors compare

to Target vectors?

Fig. 16. Flowchart of DE.

Table 7The brief summary of some of the applied DE-MPPT in the literature.


DE [126] Standard DE was used for a static objective function.[123] A modified DE technique was employed for a PV system

under partial shading. The processing time was reducedby avoiding extra measurements.

[122] A modified DE technique was employed and comparedby using a conventional HC method. The techniqueperforms well under partial shading. However, experi-mental verifications are lacking.

[128] A two-stage method based on DE in the first stage andANN in the second stage was proposed. The method iscapable under partial shading. However, it is morecomplex and lacks experimental verifications.


with several similarities to other evolutionary techniques such asDE and PSO. The main difference between RMO and other evolu-tionary techniques is that particles move around a center pointwhich is updated at each iteration. The particles spread from thecenter along the radii in straight lines based on the Vi,j vectorobtained by the following expression

( )= ( ) × =

−

= =( )

( )( ) ( )⎧

⎨⎪⎩⎪ 26

V rand V whereV

X X

ki nop j nod

0, 11, 2, ... , ; 1, 2, ... ,

i j jj

j j

, maxmax

max min

where nop and nod are the number of the particles and the searchspace dimension respectively. The coefficient k is an integer whichcan be specified appropriately. For each iteration the particles arescattered around the current center point and the particle with thebest fitness value i.e. the radial best (Rbest) and the global best(Gbest) are updated and used to determine the location of thecenter point in the search space for the next iterations, as shownby the following expression

= + × ( − )

+ × ( − ) ( )

Centre Centre Gbest Centre

Rbest Centre

C

C 27

new old1

old

2old

where C1 and C2 are coefficients which have direct impact on thevalues Gbest and Rbest, respectively.

The location of the particles around the next center point isdefined by

= × + ( )X W V Centre 28i j i jnew

, ,

One of the main differences between the RMO and PSO tech-niques is that for RMO the location and velocity of all particles arenot transferred across iterations [130]. Rather particles begin tomove from a new updated point at each iteration and as a result,less memory is required. Furthermore, using RMO the presence ofa global best vector prevents the algorithm from being trapped ata local optimum. In [131], the RMO technique was employed forGMPP tracking for a partially shaded PV system. Compared withother techniques including PSO, the technique demonstrated un-der different partial shading conditions to be faster and providemore stable tracking performance. This demonstrated perfor-mance however was limited to simulation only. To fully evaluatethe RMO-MPPT technique, experimental verification is requiredincluding considering all partial shading conditions.

3.8.2. Cuckoo search method (CS)The Cuckoo search (CS) method is based on the reproduction

strategy of some species of Cuckoo bird which are reported to laytheir eggs in the nests of other birds. This parasitic reproductionapproach is the basis of the CS optimization method. In this ap-proach, seeking the right host nest is a key step. This is similar tosearching for food, which is a random process and can be modeledusing mathematical functions. One of the common methods formodeling an animal's food seeking trajectory is the Lévy flightmodel. In the CS algorithm, the Lévy flight model is used in orderto characterise the nest seeking steps of a Cuckoo bird's re-production process. The Lévy flight model represents a randomwalk where step sizes are defined using Lévy distribution, which isdetermined by the power law, as shown in the following equation[132–134]

( )λ λé ( ) ≈ = < < ( )λ−

L u wherevy 1 , 1 3 29

For each iteration, in order to determine the new solution, theLévy flight is performed using the following expression

α λ= + ⊕ é ( ) ( )+x x L vy 30it

it1

where i is the number of samples or eggs, t is the iteration number,the product⊕indicates entry-wise multiplication, and α is the stepsize which needs to be tuned properly according to the constraintsof the optimization problem. The value of α is often defined usingthe initial step α0, and the difference between two samples (xtj andxti ), as shown by the following expression


α α= ( − ) ( )x x 310 jt

it

The operational behavior of the CS technique is similar to theHC and P&O methods. Unlike these methods however, the stepsize of the CS is based on Lévy flight and is a population-basedalgorithm. Due to Lévy flight, compared with standard PSO and DEtechniques, the CS method, has more efficient randomization andfaster convergence. Also compared with PSO method it has fewerparameters needing to be tuned and is more independent frominitial conditions. The CS-based MPPT technique was developed in[132] and was verified under certain partial shading conditions.The final experimental results demonstrated the ability of themethod to outperform the standard PSO and conventional P&Omethods.

3.8.3. Grey wolf optimization (GWO)Grey wolf optimization (GWO) is another population-based

optimization technique, based on the hierarchical leadership andthe hunting behavior of the grey wolf and recently presented in[135]. Grey wolves live in packs and have a strict social hierarchyas shown by Fig. 17. When seeking prey, grey wolves can beclassified into four types; α, β, δ, and ω, based on the fitness eva-luation of each type where α type possesses the highest fitness andω the worst. Grey wolves have three main steps in their huntingbehavior: a) searching, tracking and approaching prey; b) pursuingand encircling prey; and c) attacking prey.

Circling behavior by the wolves is a key step in the huntingprocedure and for the purpose of designing the GWO algorithm ismathematically represented by Eq. (33),

→=

→ →( ) −

→( ) ( )D C X t X t.

32p

→( + ) =

→( ) −

→ →( )X t X t A D1 . 33p

where t is the number of the iteration, Xp and X are the positionsvectors of the prey and the grey wolf, respectively; and A, C and Dare the coefficient vectors calculated by Eqs. (32), (34) and (35).

→= → → − → ( )A a r a2. . 341

→= → ( )C r2. 352

where components of a are linearly decrease from 2 to 0 acrossiterations and r1, r2 are random vectors in [0, 1]. The GWO methodis applied to MPPT where the number of participating grey wolvesrepresent the converter's duty cycles and the MPP is prey beinghunted [136].

Fig. 17. Social hierarchy of grey wolf packs (dominance decreasing from top down)[135].

3.8.4. Firefly optimization algorithm (FOA)The FOA is a population-based optimization method and si-

milar to PSO. The FOA is inspired by illuminated bugs, and ismathematically presented in [137] and [138]. Unlike PSO the FOAhas less parameters to be tuned and the particles in FOA aggregatemore closely around each optima, without rapid fluctuation likePSO. Flashing light is a key component of the population-basedbehavior of fireflies and is used to attract mating partners andpotential prey, as well as a mechanism for protective warning. Assuch, brightness is important and used to determine the newposition for particles in the search space [137,139]. For example, ifthe firefly p has lower brightness than firefly q, the new positionfor firefly p is determined by following expression

β= + ( )( − ) + −( )

+ ⎛⎝⎜

⎞⎠⎟x x r x x rand

12 36p

tpt

p q1

where, Xp and Xq represent the positions of two fireflies, r is thedistance between these two fireflies, β is the level of attractive-ness, and α is a random movement factor and is constant value inthe range of [0, 1]. As explained in [137], it is important to mentionthat a large value of α makes it possible to search for the solutionthrough a large search space whereas a smaller value of α tends tofacilitate local search [28]. The degree of attractiveness (β) in Eq.(37), can be calculated using the following expression

β β( ) = ≥ ( )γ− ( )r e n, 1; 37x0

pq n

where, γ is the absorption coefficient and controls the light in-tensity and β0 is initial attractiveness.

Using the FOA-based MPPT method, the converter duty cyclerepresents the position of the firefly and the output generatedpower of PV system is the brightness of each butterfly. The per-formance analysis of FOA-MPPT technique is carried out in [139],and it is shown that this technique outperforms the standard PSOtechnique in terms of tracking speed, tracking accuracy and dy-namic response.

4. Analytical comparison

Published research in this field indicates the difficulty of eval-uating and comparing the best MPPT approaches and techniques.In general, the final MPPT technique is selected based on the ap-plication requirements and preferences. Therefore, knowledgeabout the nature of the project and the limitations is an essentialprerequisite. In addition, the test benches, applications, and en-vironmental conditions used to verify the performance of thedesigned MPPT techniques are not similar. Therefore, a compar-ison among these methods should be fair. Compared with theconventional MPPT techniques, the intelligent approaches com-monly have lower oscillation around MPPT and higher reliabilityfor sudden changes in irradiance levels. In addition, they mostlyshow better performance in the most significant criteria. However,their behavior with regard to the main criteria, such as efficiency,reliability under PSCs, convergence speed, system independence,and steady-state oscillation, might be different. The performanceof these techniques in accordance to the accredited literature isdiscussed and presented in Table 8 and Fig. 18. In Fig. 18, the axisesA, B, C, D, E, F, refer to periodic tuning and parameter dependency,simplicity, efficiency, reliability under PSCs, system independencyand convergence speed, respectively.

According to the reviewed studies, the FLC and ANN methods intheir original forms are incapable of tracking the global maximumpoints under PSCs. However, they obtain satisfactory results undernormal conditions. The most significant advantage of the ANNmethod is its independence from detailed information of the PV

Table 8Comparison of different MPPT techniques.

Criteria ANN FLC PSO ACO DE GA GWO/FOA RMO/CS

Convergence speed Average Average Fast Fast Fast Fast Medium Very fastSystem independency Poor Poor High High High High High HighAbility to track under PSCs Poor Poor High High Average high Average high High HighAbility to Track under normal condition High High High High High High High HighEfficiency Poor-in PSCs Poor- in PSCs High High High High High Very HighAlgorithm complexity Simple Medium Simple Simple Simple Complex Medium MediumImplementation cost and complexity High High Average -low Average low Average low Average low Average low Average lowPeriodic tuning Yes Yes No No No No No NoDependency of the initial design High High Medium Medium Low Medium Medium MediumOscillation around MPP No No No No No No No No


system. However, its reliability is largely affected by the trainingprocess, which makes the system exclusive for each PV system.Therefore, the ANN method cannot operate for another PV systemunless it undergoes a new training process. This problem may alsooccur when the characteristics of the array change due to agingor degradation. The main advantages of FLC are its system in-dependence, ease of implementation, and satisfactory performance

Fuzzy

PSO

DE

D

A

Conventional Methods

ALow to High LD

A

y

ALow to High L

ConventioMethod

ConventioMethods


RMO/CS

D

A

Low to High


Fig. 18. Comparison of new and conventional MPP

under FICC. However, the computational cost of the system fordesigning the fuzzification rule base and defuzzification processincreases. Compared with other AI-based MPP techniques, the FLCrequires more perception and comprehension of the PV system bythe designer to design the different FLC parameters, such as rulesand membership functions. Therefore, this method can be con-sidered more system dependent than others. System independence,

GA

D D

Aow to High

D D

Aow to High Low to High

Low to High

nal s



nal

GWO/FOA

D

A

Low to High


T methods according to six main parameters.


speed, and decision making based on approximate values make thefuzzy logic controller one of the most suitable techniques forcomplex engineering challenges. Thus, this method can be con-sidered one of the more reliable options for finding the MPP in thenonlinear characteristics of PV systems under normal conditionsand some PSCs. However, this technique highly depends on de-signer knowledge in the different steps of the algorithms. In addi-tion, the method needs extensive computation in the fuzzification,rule base, and defuzzification stages.

All the PSO-based methods, both original and modified ver-sions, are suitable to operate under PSC. The methods are mainlyrelated to their heuristic behavior for exploring the search spacethat comprise multiple maximum power points. The main pro-blems of the PSO method are implementation complexity, poordynamic response due to the interruption for measurement pro-cess, long tracking process as a result of undefined initial points,and significant dependency on the coefficient designs and initialconditions. The modified version of the PSO method solved theinitialization problems; however, it does not guarantee that theMPP under any PSCs can be tracked, and the initial points are setclose the global MPPs. Overall, PSO has a fast convergence speed,high sensitivity atmospheric change, low hardware implementa-tion cost, and high efficiency with no oscillation around MPP.

ACO can also track the MPP under both normal and PSC. UnlikePSO, ACO is not dependent on the design of the initial location ofthe particles. In addition, it is capable of tracking MPP in a systemwith minimized sensors, which results in a more cost-effectivesystem. However, theoretically, this condition can be similar for allthe metaheuristic approaches, such as PSO, DE, and GA. In general,ACO comprises three factors: the positive feedback mechanismthat increases the probability of detecting optimal solutions ininitial iterations, the distributed computation to ensure that thealgorithm is not involved in the local optimum, and the greedysearch that helps the algorithm find the optimal solution withhigher convergence speed [108]. Overall, the most significant ad-vantages of the ACO over other AI-based methods are system in-dependence, high convergence speed, and initial location in-dependence of particles. However, these advantages might in-crease the computational burden because of complex calculations.In addition, due to the lack of research in this area, the reliabilityand robustness of the technique under different PSCs have notbeen experimentally verified.

The GA and DE techniques, which have almost similar concepts,can track the global MPP under PSCs because of their ability tosolve multi-objective problems. Both algorithms are system in-dependent, efficient, have no need for periodic tuning, and haveno oscillation around MPP. Moreover, the advantages of the DEalgorithm are simplicity because of the few required controlparameters, fast convergence speed, and initial location in-dependence. However, experimental verifications for purely usingeither of these methods under PSC conditions are lacking. Unlikethe PSO and ACO theories, the methods do not remember previousmovements and locations that the particles have experiencedthroughout the program. Therefore, the algorithm is more likely tobe stuck in the local optimum.

Of the four emerging techniques, RMO and CS appear to havesimilar functionality and address similar limitations of the stan-dard PSO technique. In RMO particles explore a fraction of mainsearch space therefore achieving faster convergence speed com-pared to other MPPT methods. The CS MPPT technique also has avery fast convergence speed compared to other MPPT techniques,achieved using Lévy flight during its operation. Both techniquesrequire only a few parameters to be tuned however the CS tech-nique has fairly complex operation compared with other softcomputing based MPPT methods and RMO has not been experi-mentally tested. GWO and FAO techniques have demonstrated

similar performance under partial shading conditions. The con-vergence speed towards the GMPP for both algorithms is an im-provement upon common AI-techniques, and they have lesscomplex structure compared with RMO and CS algorithms. How-ever GWO or FAO based MPPT unit results in output oscillationduring tracking period compared with RMO and CS algorithms.Therefore, in the applications which are very sensitive to outputpower fluctuations, RMO or CS technique are preferred.

5. Conclusion

In view of the importance of control strategies in the overallefficiency of the PV systems, this paper focused on the differentapproaches in tracking the MPP of PV systems. Given the draw-backs of conventional MPPT methods, such as system in-dependence, high oscillation around MPP, and deficiency underPSCs, six common and four emerging AI-based MPPT approacheswere reviewed in this paper. The concept, structure, sequentialsteps, and the state of the art of each method of tracking the MPPunder PSCs have been thoroughly presented. To comprehend thesignificance of the environmental conditions on the performanceof the PV system, an extensive analysis on the behavior of PVsystems under normal and PSCs was carried out in advance. Inaddition, the latest studies that conducted performance analysesof each method under different conditions were discussed andanalyzed. This paper also presented the different hybrid methodsthat used one of the covered AI methods.

The final evaluation was carried out to evaluate the complexity,ability to track GMPP under PSC, sensitivity to atmosphericchanges, convergence speed, cost, system independence, effi-ciency, periodic training or coefficient tuning process, and oscil-lation around each method of MPP. The comparison was con-ducted mainly according to the concept of the method, as thevalidations reported in the studies were performed on differentsystems and under different conditions. This study is not limited toreviewing the performance of the recent methods and the step-by-step procedures and structures of each method; it also com-prehensively covered the required references. The findings of thispaper will help researchers and engineers who work with PV-based power systems to understand the basics of each method andselect the appropriate one according to project requirements to beimplemented in their MPPT systems to fulfill project objectives.

References

[1] Zahedi A. Solar photovoltaic (PV) energy; latest developments in the buildingintegrated and hybrid PV systems. Renew Energy 2006;31:711–8.

[2] Zahedi A. Australian renewable energy progress. Renew Sustain Energy Rev2010;14:2208–13.

[3] Solangi KH, Islam MR, Saidur R, Rahim NA, Fayaz H. A review on global solarenergy policy. Renew Sustain Energy Rev 2011;15:2149–63.

[4] Salam Z, Ahmed J, Merugu BS. The application of soft computing methods forMPPT of PV system: a technological and status review. Appl Energy2013;107:135–48.

[5] Hussein KH, Muta I, Hoshino T, Osakada M. Maximum photovoltaic powertracking: an algorithm for rapidly changing atmospheric conditions. IEE ProcGener Transm Distrib 1995;142:59–64.

[6] Hua C, Lin J, Shen C. Implementation of a DSP-controlled photovoltaic systemwith peak power tracking. IEEE Trans Ind Electron 1998;45:99–107.

[7] Hua C, Lin J. An on-line MPPT algorithm for rapidly changing illuminations ofsolar arrays. Renew Energy 2003;28:1129–42.

[8] Femia N, Petrone G, Spagnuolo G, Vitelli M. Optimization of perturb andobserve maximum power point tracking method. IEEE Trans Power Electron2005;20:963–73.

[9] Ahmed J, Salam Z. An improved perturb and observe (P&O) maximum powerpoint tracking (MPPT) algorithm for higher efficiency. Appl Energy2015;150:97–108.

[10] Weidong X, Dunford WG. A modified adaptive hill climbing MPPT methodfor photovoltaic power systems. Power electronics specialists conference,

http://refhub.elsevier.com/S1364-0321(16)30284-2/sbref1

































2004 PESC 04 2004 IEEE 35th Annual, vol. 3; 2004, p. 1957–1963.[11] Shimizu T, Hashimoto O, Kimura G. A novel high-performance utility-interactive

photovoltaic inverter system. IEEE Trans Power Electron 2003;18:704–11.[12] Ma L, Sun Y, Lin Y, Bai Z, Tong L, Song J. A high performance MPPT control

method. Mater Renew Energy Environ (ICMREE) 2011:195–9.[13] Fangrui L, Yong K, Yu Z, Shanxu D. Comparison of P&O and hill climbing

MPPT methods for grid-connected PV converter. Ind Electron Appl2008:804–7.

[14] Ahmed A, Li R, Bumby J. Perturbation parameters design for hill climbingMPPT techniques. Ind Electron (ISIE) 2012:1819–24.

[15] Yeong-Chan K, Tsorng-Juu L, Jiann-Fuh C. Novel maximum-power-point-tracking controller for photovoltaic energy conversion system. IEEE Trans IndElectron 2001;48:594–601.

[16] Safari A, Mekhilef S. Simulation and hardware implementation of incre-mental conductance MPPT with direct control method using Cuk converter.IEEE Trans Ind Electron 2011;58:1154–61.

[17] Qiang M, Mingwei S, Liying L, Guerrero JM. A novel improved variable step-size incremental-resistance MPPT method for PV systems. IEEE Trans IndElectron 2011;58:2427–34.

[18] Patel H, Agarwal V. Maximum power point tracking scheme for PV systemsoperating under partially shaded conditions. IEEE Trans Ind Electron2008;55:1689–98.

[19] Kobayashi K, Takano I, Sawada Y. A study of a two stage maximum powerpoint tracking control of a photovoltaic system under partially shaded in-solation conditions. Solar Energy Mater Solar Cells 2006;90:2975–88.

[20] Tey Kok S, Mekhilef S. A fast-converging MPPT technique for photovoltaicsystem under fast-varying solar irradiation and load resistance. IEEE TransInd Inform 2015;11:176–86.

[21] Tey KS, Mekhilef S. Modified incremental conductance MPPT algorithm tomitigate inaccurate responses under fast-changing solar irradiation level.Solar Energy 2014;101:333–42.

[22] Yi-Hwa L, Shyh-Ching H, Jia-Wei H, Wen-Cheng L. A particle swarm opti-mization-based maximum power point tracking algorithm for PV systemsoperating under partially shaded conditions. IEEE Trans Energy Convers2012;27:1027–35.

[23] Tsang KM, Chan WL. Maximum power point tracking for PV systems underpartial shading conditions using current sweeping. Energy Convers Manag2015;93:249–58.

[24] Kok Soon T, Mekhilef S. Modified incremental conductance algorithm forphotovoltaic system under partial shading conditions and load variation.IEEE Trans Ind Electron 2014;61:5384–92.

[25] Reza Reisi A, Hassan Moradi M, Jamasb S. Classification and comparison ofmaximum power point tracking techniques for photovoltaic system: a re-view. Renew Sustain Energy Rev 2013;19:433–43.

[26] Kamarzaman NA, Tan CW. A comprehensive review of maximum powerpoint tracking algorithms for photovoltaic systems. Renew Sustain EnergyRev 2014;37:585–98.

[27] Liu Y-H, Chen J-H, Huang J-W. A review of maximum power point trackingtechniques for use in partially shaded conditions. Renew Sustain Energy Rev2015;41:436–53.

[28] Bidram A, Davoudi A, Balog RS. Control and circuit techniques to mitigatepartial shading effects in photovoltaic arrays. IEEE J Photovoltaics2012;2:532–46.

[29] Eltawil MA, Zhao Z. MPPT techniques for photovoltaic applications. RenewSustain Energy Rev 2013;25:793–813.

[30] Bhatnagar P, Nema RK. Maximum power point tracking control techniques:state-of-the-art in photovoltaic applications. Renew Sustain Energy Rev2013;23:224–41.

[31] De Brito MAG, Galotto L, Sampaio LP, De Azevedo Melo G, Canesin CA.Evaluation of the main MPPT techniques for photovoltaic applications. IEEETrans Ind Electron 2013;60:1156–67.

[32] Esram T, Chapman PL. Comparison of photovoltaic array maximum powerpoint tracking techniques. IEEE Trans Energy Convers EC 2007;22:439.

[33] Salas V, Olías E, Barrado A, Lázaro A. Review of the maximum power pointtracking algorithms for stand-alone photovoltaic systems. Solar EnergyMater Solar Cells 2006;90:1555–78.

[34] Ishaque K, Salam Z. A review of maximum power point tracking techniquesof PV system for uniform insolation and partial shading condition. RenewSustain Energy Rev 2013;19:475–88.

[35] Seyedmahmoudian M, Mekhilef S, Rahmani R, Yusof R, Renani ET. Analyticalmodeling of partially shaded photovoltaic systems. Energies 2013;6:128–44.

[36] Seyedmahmoudian M, Rahmani R, Mekhilef S, Maung Than Oo A, StojcevskiA, Tey Kok S, et al. Simulation and hardware implementation of new max-imum power point tracking technique for partially shaded PV system usinghybrid DEPSO method. IEEE Trans Sustain Energy 2015;6:850–62.

[37] Mellit A, Kalogirou SA. Artificial intelligence techniques for photovoltaicapplications: a review. Prog Energy Combust Sci 2008;34:574–632.

[38] Kalogirou SA. Artificial intelligence for the modeling and control of com-bustion processes: a review. Prog Energy Combust Sci 2003;29:515–66.

[39] Hong-Hee L, Le Minh P, Phan Quoc D, Nguyen Truong Dan V, Le Dinh K. Thenew maximum power point tracking algorithm using ANN-based solar PVsystems. In: TENCON 2010 – 2010 IEEE Region 10 Conference; 2010. p. 2179–84.

[40] Liu Y-H, Liu C-L, Huang J-W, Chen J-H. Neural-network-based maximumpower point tracking methods for photovoltaic systems operating under fastchanging environments. Solar Energy 2013;89:42–53.

[41] Hiyama T, Kouzuma S, Imakubo T, Ortmeyer T. Evaluation of neural network

based real time maximum power tracking controller for PV system. IEEETrans Energy Convers 1995;10:543–8.

[42] Al-Amoudi A, Zhang L. Application of radial basis function networks for solar-array modelling and maximum power-point prediction. In: IEE Proceedings ofthe Generation, Transmission and Distribution, IET; 2000. p. 310–6.

[43] Veerachary M, Yadaiah N. ANN based peak power tracking for PV suppliedDC motors. Solar Energy 2000;69:343–50.

[44] Bahgat A, Helwa N, Ahmad G, El Shenawy E. Maximum power point trackingcontroller for PV systems using neural networks. Renew Energy2005;30:1257–68.

[45] Alabedin AZ, El-Saadany E, Salama M. Maximum power point tracking forPhotovoltaic systems using fuzzy logic and artificial neural networks. In:Proceedings of the 2011 IEEE on power and energy society general meeting,IEEE; 2011. p. 1–9.

[46] Islam MA, Kabir MA. Neural network based maximum power point trackingof photovoltaic arrays. In: Proceedings of the TENCON 2011 – 2011 IEEERegion 10 Conference; 2011. p. 79–82.

[47] Jie L, Ziran C. Research on the MPPT algorithms of photovoltaic system basedon PV neural network. In: Proceedings of the 2011 Chinese control and de-cision conference (CCDC), IEEE; 2011. p. 1851–4.

[48] Veerachary M, Senjyu T, Uezato K. Neural-network-based maximum-power-point tracking of coupled-inductor interleaved-boost-converter-supplied PVsystem using fuzzy controller. IEEE Trans Ind Electron 2003;50:749–58.

[49] Mellit A, Kalogirou SA. MPPT-based artificial intelligence techniques forphotovoltaic systems and its implementation into field programmable gatearray chips: Review of current status and future perspectives. Energy2014;70:1–21.

[50] Laudani A, Fulginei FR, Salvini A, Lozito GM, Mancilla-David F. Im-plementation of a neural MPPT algorithm on a low-cost 8-bit micro-controller. In: Proceedings of the 2014 international symposium on powerelectronics, electrical drives, automation and motion (SPEEDAM); 2014. p.977–81.

[51] Amrouche B, Belhamel M, Guessoum A. Maximum power point trackingacceleration by using modified P&O method for photovoltaic systems. In:Proceedings of the Second international congress on environment and re-newable energies, Mahdia, Tunisia; 2006. p. 6–8.

[52] Aashoor FAO, Robinson FVP. A variable step size perturb and observe algo-rithm for photovoltaic maximum power point tracking. In: Proceedings ofthe 2012 47th international universities power engineering conference(UPEC); 2012. p. 1–6.

[53] Femia N, Petrone G, Spagnuolo G, Vitelli M. Optimizing duty-cycle pertur-bation of P&O MPPT technique. In: Power electronics specialists conference,2004 PESC 04 2004 IEEE 35th Annual, vol. 3, 2004. p. 1939–44.

[54] Punitha K, Devaraj D, Sakthivel S. Artificial neural network based modified in-cremental conductance algorithm for maximum power point tracking in pho-tovoltaic system under partial shading conditions. Energy 2013;62:330–40.

[55] Xu J, Shen A, Yang C, Rao W, Yang X. ANN based on IncCond algorithm forMPP tracker. In: Proceedings of the 2011 sixth international conference onbio-inspired computing: theories and applications (BIC-TA), IEEE; 2011. p.129–34.

[56] Ramaprabha R, Gothandaraman V, Kanimozhi K, Divya R, Mathur BL. Max-imum power point tracking using GA-optimized artificial neural network forSolar PV system. In: Proceedings of the 2011 1st international conference onelectrical energy systems (ICEES); 2011. p. 264–8.

[57] Ishaque K, Abdullah SS, Ayob SM, Salam Z. Single input fuzzy logic controllerfor unmanned underwater vehicle. J Intell Robot Syst 2010;59:87–100.

[58] Chian-Song C. T-S Fuzzy maximum power point tracking control of solarpower generation systems. IEEE Trans Energy Convers 2010;25:1123–32.

[59] Syafaruddin Karatepe E, Hiyama T. Artificial neural network-polar co-ordinated fuzzy controller based maximum power point tracking controlunder partially shaded conditions. IET Renew Power Gener 2009;3:239–53.

[60] Fullér R. Neural fuzzy systems; 1995.[61] Ross TJ. Fuzzy logic with engineering applications. John Wiley & Sons; 2009.[62] Sivanandam S, Sumathi S, Deepa S. Introduction to fuzzy logic using MA-

TLAB. Springer; 2007.[63] Masoum M, Sarvi M. A new fuzzy-based maximum power point tracker for

photovoltaic applications. Iran J Electr Electron Eng 2005;1:28–35.[64] Alajmi BN, Ahmed KH, Finney SJ, Williams BW. A maximum power point

tracking technique for partially shaded photovoltaic systems in microgrids.IEEE Trans Ind Electron 2013;60:1596–606.

[65] Kottas TL, Boutalis YS, Karlis AD. New maximum power point tracker for PVarrays using fuzzy controller in close cooperation with fuzzy cognitive net-works. IEEE Trans Energy Convers 2006;21:793–803.

[66] Chung-Yuen W, Duk-Heon K, Sei-Chan K, Won-Sam K, Hack-Sung K. A newmaximum power point tracker of photovoltaic arrays using fuzzy con-troller. In: Power Electronics Specialists Conference, PESC '94 Record, 25thAnnual IEEE, vol.1, 1994. p. 396–403.

[67] Simoes MG, Franceschetti NN, Friedhofer M. A fuzzy logic based photovoltaicpeak power tracking control. Ind Electron 1998;1:300–5.

[68] Ben Salah C, Ouali M. Comparison of fuzzy logic and neural network inmaximum power point tracker for PV systems. Electr Power Syst Res2011;81:43–50.

[69] Algazar MM, Al-monier H, El-halim HA, Salem MEEK. Maximum power pointtracking using fuzzy logic control. Int J Electr Power Energy Syst 2012;39:21–8.

[70] Liu C-L, Chen J-H, Liu Y-H, Yang Z-Z. An asymmetrical fuzzy-logic-control-based MPPT algorithm for photovoltaic systems. Energies 2014;7:2177–93.














































































































































































[71] Adly M, Besheer AH. A meta-heuristics search algorithm as a solution forenergy transfer maximization in stand-alone photovoltaic systems. Int JElectr Power Energy Syst 2013;51:243–54.

[72] Larbes C, Aït Cheikh SM, Obeidi T, Zerguerras A. Genetic algorithms opti-mized fuzzy logic control for the maximum power point tracking in photo-voltaic system. Renew Energy 2009;34:2093–100.

[73] Letting LK, Munda JL, Hamam A. Particle swarm optimized TS fuzzy logiccontroller for maximum power point tracking in a photovoltaic system. In:IPEC, 2010 conference proceedings: IEEE; 2010. p. 89–94.

[74] Radjai T, Rahmani L, Mekhilef S, Gaubert JP. Implementation of a modified in-cremental conductance MPPT algorithm with direct control based on a fuzzyduty cycle change estimator using dSPACE. Solar Energy 2014;110:325–37.

[75] Messai A, Mellit A, Guessoum A, Kalogirou SA. Maximum power pointtracking using a GA optimized fuzzy logic controller and its FPGA im-plementation. Solar Energy 2011;85:265–77.

[76] Al Nabulsi A, Dhaouadi R. Efficiency optimization of a dsp-based standalonePV system using fuzzy logic and dual-MPPT control. IEEE Trans Ind Inform2012;8:573–84.

[77] Guenounou O, Dahhou B, Chabour F. Adaptive fuzzy controller based MPPTfor photovoltaic systems. Energy Convers Manag 2014;78:843–50.

[78] Liu CL, Chen JH, Liu YH, Yang ZZ. An asymmetrical fuzzy-logic-control-basedMPPT algorithm for photovoltaic systems. Energies 2014;7:2177–93.

[79] Rajesh R, Mabel MC. Efficiency analysis of a multi-fuzzy logic controller forthe determination of operating points in a PV system. Solar Energy2014;99:77–87.

[80] Farhat M, Barambones O, Sbita L. Efficiency optimization of a DSP-basedstandalone PV system using a stable single input fuzzy logic controller. Re-new Sustain Energy Rev 2015;49:907–20.

[81] Chikh A, Chandra A. An optimal maximum power point tracking algorithmfor PV Systems with climatic parameters estimation. IEEE Trans SustainEnergy 2015;6:644–52.

[82] Young-Hyok J, Doo-Yong J, Jun-Gu K, Jae-Hyung K, Tae-Won L, Chung-YuenW. A real maximum power point tracking method for mismatching com-pensation in PV array under partially shaded conditions. IEEE Trans PowerElectron 2011;26:1001–9.

[83] Ramaprabha R, Balaji M, Mathur BL. Maximum power point tracking ofpartially shaded solar PV system using modified Fibonacci search methodwith fuzzy controller. Int J Electr Power Energy Syst 2012;43:754–65.

[84] Eberhart RC, Kennedy J. A new optimizer using particle swarm theory. In:Proceedings of the sixth international symposium on micro machine andhuman science: New York, NY; 1995. p. 39–43.

[85] Eberhart RC, Shi Y, Kennedy J. Swarm intelligence. Elsevier; 2001.[86] James K, Russell E. Particle swarm optimization. In: Proceedings of the 1995

IEEE International conference on neural networks; 1995. p. 1942–8.[87] Seo JH, Im CH, Heo CG, Kim JK, Jung HK, Lee CG. Multimodal function opti-

mization based on particle swarm optimization. IEEE Trans Magn2006;42:1095–8.

[88] Phimmasone V, Endo T, Kondo Y, Miyatake M. Improvement of the max-imum power point tracker for photovoltaic generators with particle swarmoptimization technique by adding repulsive force among agents. Electr MachSyst 2009:1–6.

[89] Miyatake M, Veerachary M, Toriumi F, Fujii N, Ko H. Maximum power pointtracking of multiple photovoltaic arrays: a PSO approach. IEEE Trans AerospElectron Syst 2011;47:367–80.

[90] Seyedmahmoudian M, Mekhilef S, Rahmani R, Yusof R, Asghar Shojaei A.Maximum power point tracking of partial shaded photovoltaic array usingan evolutionary algorithm: A particle swarm optimization technique. J Re-new Sustain Energy 2014:6.

[91] Kaewkamnerdpong B, Bentley PJ. Perceptive particle swarm optimisation: aninvestigation. In: Proceedings 2005 IEEE Swarm Intelligence Symposium,2005 SIS 2005; 2005. p. 169–76.

[92] Roy Chowdhury S, Saha H. Maximum power point tracking of partiallyshaded solar photovoltaic arrays. Solar Energy Mater Solar Cells2010;94:1441–7.

[93] Ishaque K, Salam Z, Amjad M, Mekhilef S. An improved particle swarm op-timization (PSO)-based MPPT for PV with reduced steady-state oscillation.IEEE Trans Power Electron 2012;27:3627–38.

[94] Ishaque K, Salam Z. A deterministic particle swarm optimization maximumpower point tracker for photovoltaic system under partial shading condition.IEEE Trans Ind Electron 2013;60:3195–206.

[95] Lian KL, Jhang JH, Tian IS. A maximum power point tracking method basedon perturb-and-observe combined with particle swarm optimization. IEEE JPhotovoltaics 2014;4:626–33.

[96] Dorigo M, Gambardella LM. Ant colony system: a cooperative learning ap-proach to the traveling salesman problem. IEEE Trans Evolut Comput1997;1:53–66.

[97] Shen Q, Jiang J-H, Tao J-c, Shen G-l, Yu R-Q. Modified ant colony optimizationalgorithm for variable selection in QSAR modeling: QSAR studies of cy-clooxygenase inhibitors. J Chem Inf Model 2005;45:1024–9.

[98] Dorigo M, Birattari M, Stutzle T. Ant colony optimization. IEEE Comput In-telligence Mag 2006;1:28–39.

[99] Socha K, Dorigo M. Ant colony optimization for continuous domains. Eur JOper Res 2008;185:1155–73.

[100] Dorigo M, Maniezzo V, Colorni A. Ant system: optimization by a colony of co-operating agents. IEEE Trans Syst Man Cybern Part B: Cybern 1996;26:29–41.

[101] Rahmani R, Yusof R, Seyedmahmoudian M, Mekhilef S. Hybrid technique of

ant colony and particle swarm optimization for short term wind energyforecasting. J Wind Eng Ind Aerodyn 2013;123:163–70.

[102] Dorigo M, Caro G, Gambardella L. Ant algorithms for discrete optimization.Artif Life 1999;5:137–72.

[103] Corne D, Dorigo M, Glover F, Dasgupta D, Moscato P, Poli R, et al. New ideasin optimization. UK: McGraw-Hill Ltd; 1999.

[104] Yu L, Liu K, Li K. Ant colony optimization in continuous problem. Front MechEng China 2007;2:459–62.

[105] Chen L, Shen J, Qin L, Chen H. An improved ant colony algorithm in con-tinuous optimization. J Syst Sci Syst Eng 2003;12:224–35.

[106] Kıran MS, Gündüz M. A recombination-based hybridization of particle swarmoptimization and artificial bee colony algorithm for continuous optimizationproblems. Appl Soft Comput 2013;13:2188–203.

[107] Toksari MD. Ant colony optimization for finding the global minimum. ApplMath Comput 2006;176:308–16.

[108] Jiang LL, Maskell DL, Patra JC. A novel ant colony optimization-based max-imum power point tracking for photovoltaic systems under partially shadedconditions. Energy Build 2013;58:227–36.

[109] Besheer A, Adly M. Ant colony system based PI maximum power pointtracking for stand alone photovoltaic system. In: Proceedings of the 2012IEEE International Conference on Industrial Technology (ICIT), IEEE; 2012. p.693–8.

[110] Adly M, Besheer AH. A meta-heuristics search algorithm as a solution forenergy transfer maximization in stand-alone photovoltaic systems. Int JElectr Power Energy Syst 2013;51:243–54.

[111] Holland JH. Adaptation in natural and artificial systems: an introductoryanalysis with applications to biology, control, and artificial intelligence. UMichigan Press; 1975.

[112] Shaiek Y, Ben Smida M, Sakly A, Mimouni MF. Comparison between con-ventional methods and GA approach for maximum power point tracking ofshaded solar PV generators. Solar Energy 2013;90:107–22.

[113] Daraban S, Petreus D, Morel C. A novel global MPPT based on genetic algo-rithms for photovoltaic systems under the influence of partial shading. In:Proceedings of the IECON 2013-39th Annual Conference of the IEEE IndustrialElectronics Society, IEEE; 2013. p. 1490–5.

[114] Goldberg DE, Holland JH. Genetic algorithms and machine learning. MachLearn 1988;3:95–9.

[115] Ramaprabha R MB. Genetic algorithm based maximum power point trackingfor partially shaded solar photovoltaic array IntJResRevInfSci (IJRRIS).

[116] Mohajeri HR, Moghaddam MP, Shahparasti M, Mohamadian M. Developmenta new algorithm for maximum power point tracking of partially shadedphotovoltaic arrays. In: Proceedings of the 2012 20th Iranian conference onelectrical engineering (ICEE), IEEE; 2012. p. 489–94.

[117] Daraban S, Petreus D, Morel C. A novel MPPT (maximum power pointtracking) algorithm based on a modified genetic algorithm specialized ontracking the global maximum power point in photovoltaic systems affectedby partial shading. Energy 2014.

[118] Messai A, Mellit A, Guessoum A, Kalogirou S. Maximum power point trackingusing a GA optimized fuzzy logic controller and its FPGA implementation.Solar Energy 2011;85:265–77.

[119] Kulaksız AA, Akkaya R. A genetic algorithm optimized ANN-based MPPT al-gorithm for a stand-alone PV system with induction motor drive. Solar En-ergy. 2012;86:2366–75.

[120] Storn R, Price K. Differential evolution – a simple and efficient Heuristic forglobal optimization over continuous spaces. J Global Optim 1997;11:341–59.

[121] R Storn, K. Price Differential evolution-a simple and efficient adaptive schemefor global optimization over continuous spaces: ICSI Berkeley; 1995.

[122] Mohammad Faridun Naim T, Shahrin MA, Zainal S, Mohd Sazli S. Evolu-tionary based maximum power point tracking technique using differentialevolution algorithm; 2013.

[123] Tey KS, Mekhilef S, Yang H-T, Chuang M-K. A differential evolution basedMPPT method for photovoltaic modules under partial shading conditions. IntJ Photoenergy 2014 2014.

[124] Okdem S. A simple and global optimization algorithm for engineering pro-blems: differential evolution algorithm. Turk J Electr Eng 2004:12.

[125] Price KV. Differential evolution: a fast and simple numerical optimizer. FuzzyInformation Processing Society, 1996 NAFIPS. Berkeley, CA, USA: IEEE; 1996.p. 524–7.

[126] Price KV, Storn RM, Lampinen JA. Differential evolution: a practical approachto global optimization.New York: Springer-Verlag New York Inc; 2005.

[127] Taheri H, Salam Z, Ishaque K. A novel maximum power point tracking controlof photovoltaic system under partial and rapidly fluctuating shadow condi-tions using differential evolution. In: Proceedings of the 2010 IEEE sympo-sium onindustrial electronics and applications (ISIEA), IEEE; 2010. p. 82–7.

[128] Tajuddin MFN, Ayob SM, Salam Z. Tracking of maximum power point inpartial shading condition using differential evolution (DE). Power Energy((PECon)) 2012:384–9.

[129] Sheraz M, Abido MA. An efficient MPPT controller using differential evolutionand neural network. Power Energy ((PECon)) 2012:378–83.

[130] Rahmani R, Yusof R. A new simple, fast and efficient algorithm for globaloptimization over continuous search-space problems: radial movement op-timization. Appl Math Comput 2014;248:287–300.

[131] Seyedmahmoudian M, Horan B, Rahmani R, Maung Than Oo A, Stojcevski A.Efficient photovoltaic system maximum power point tracking using a newtechnique. Energies 2016;9:147.

[132] Ahmed J, Salam Z. A maximum power point tracking (MPPT) for PV system

















































































































































































using Cuckoo search with partial shading capability. Appl Energy2014;119:118–30.

[133] Teuschl Y, Taborsky B, Taborsky M. How do cuckoos find their hosts? The roleof habitat imprinting Animal Behav 1998;56:1425–33.

[134] Amir A, Amir A, Selvaraj J, Rahim NA. Study of the MPP tracking algorithms:focusing the numerical method techniques. Renew Sustain Energy Rev2016;62:350–71.

[135] Mirjalili S, Mirjalili SM, Lewis A. Grey Wolf Optimizer. Adv Eng Softw2014;69:46–61.

[136] Mohanty S, Subudhi B, Ray PK. A new MPPT design using grey wolf opti-mization technique for photovoltaic system under partial shading conditions.IEEE Trans Sustain Energy 2016;7:181–8.

[137] Yang X-S. Firefly algorithms for multimodal optimization. Stochastic algo-rithms: foundations and applications, Springer; 2009. p. 169–78.

[138] Yang X-S. Nature-inspired metaheuristic algorithms. Luniver Press; 2010.[139] Sundareswaran K, Peddapati S, Palani S. MPPT of PV systems under partial

shaded conditions through a colony of flashing fireflies. IEEE Trans EnergyConvers 2014;29:463–72.























Renewable and Sustainable Energy Reviews - UMEXPERT · PDF fileState of the art artiﬁcial...

Documents

Transcript of Renewable and Sustainable Energy Reviews - UMEXPERT · PDF fileState of the art artiﬁcial...