Renewable and Sustainable Energy Reviews 74 (2017) 1147–1169

Contents lists available at ScienceDirect

Renewable and Sustainable Energy Reviews

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

A new MPPT scheme based on a novel fuzzy approach MARK

⁎
M. Nabipour , M. Razaz, S.GH Seifossadat, S.S. Mortazavi
Department of Electrical Engineering, Faculty of Engineering, Shahid Chamran University of Ahvaz, Ahvaz, Iran

A R T I C L E I N F O A BS T RAC T

Keywords: In this paper, after defining Maximum Power Point Tracking (MPPT) control expectations, with the aim of
MPPT finding the optimum routine in fulfilling these expectations, a review over the available MPPT control methods
PV is presented. Throughout the review, by comparing conventional MPPT routines in terms of accomplishing
Membership functions tuning defined control objectives, the necessity of designing a new MPPT control scheme based on adaptive fuzzy logic
Adaptive
is expressed. Based on the conducted review, a new routine to optimize the MPPT performance of a Photovoltaic
Fuzzy controller
(PV)-setup and to fulfill all the MPPT control requirements is proposed. The optimization is performed in
tracking the Maximum Power Point (MPP) of the PV-module by a Boost-converter using an “antecedent-
consequent adaptive” indirect fuzzy-based MPPT scheme. The fuzzy-based scheme is tuned online using a novel
computationally light membership function tuning routine, where the antecedent and consequent membership
functions are tuned synchronously. As a result, a fast, smooth and computationally light MPPT controller is
proposed. In this regard, the presented MPPT scheme tuned using the proposed novel tuning routine is
compared with conventional direct and indirect fuzzy-based MPPT schemes, showing superiority of the
proposed MPPT routine over conventional schemes.

1. Introduction main goal in MPPT is operating the PV-setup at a point where the
highest efficiency level possible is obtained. In this regard, various
With continuously growing concerns about fossil fuels and their routines were implemented to track the MPP where Perturb and
effects on environments, the attentions are attracted to renewable Observe (P & O) [3], incremental conductance (IC) [4] and Hill
energy sources as replacements. Among different types of renewable Climbing (HC) [5] could be counted as the most common conventional
energies, because of easy implementation, low maintenance and MPPT routines. Easy implementation and high reliability in nominal
constantly dropping PV-module prices, the solar energy has been cases could be outlined as advantages of these routines. Yet, perfor-
considered as a suitable replacement for conventional energy sources. mance of the mentioned schemes deviate from desirable expectations
As a result, PV plants have been utilized globally in various regions [1]. when operating in realistic environmental conditions consisting of
Despite all merits and advances in PV technology, higher initial sudden variations in irradiance and temperature levels. As a result of
investment cost is imposed to the consumers compared to its conven- these deviations, power oscillations and losses increase, resulting in
tional fossil fuel counterpart [1]. Moreover, highly nonlinear charac- lower system efficiency [6].
teristics of PV-arrays caused by dependence of PV-modules to the On the other hand, it has been acknowledged that the Artificial
surrounding ambient conditions have limited the global utilization of Intelligence (AI) approaches express superior performance compared
solar energy. To overcome these drawbacks, two solutions have been to conventional MPPT methods, in terms of speed, accuracy and
presented by the researchers. Improving the silicon semiconductors efficiency [7].
employed in the PV-modules has been considered as the first solution. Considering drawbacks of conventional MPPT schemes, nonlinear
Therefore, new semiconductors were tested which improved the and time variant nature of MPPT tasks, recent advances in implemen-
performance of PV-setups. Yet, the improved modules were available tation of AI methods due to recent trends in micro controller
at a very high price limiting the utilization of such modules to certain technology and superior performance of AI methods, AI schemes have
fields such as space and military studies [2]. been considered as a reliable alternative for conventional MPPT
In the other solution presented by the researchers improving the methods [ 9]. In fact, the desire to utilize AI lies in the approach that
performance of PV-modules in tracking the MPP, is considered. The is intended to build applications that possess some intelligence, or

⁎
Corresponding author.
E-mail addresses: [email protected] (M. Nabipour), [email protected] (M. Razaz), [email protected] (S.G. Seifossadat),
[email protected] (S.S. Mortazavi).

http://dx.doi.org/10.1016/j.rser.2017.02.054
Received 31 October 2016; Received in revised form 20 January 2017; Accepted 12 February 2017
1364-0321/ © 2017 Elsevier Ltd. All rights reserved.
M. Nabipour et al. Renewable and Sustainable Energy Reviews 74 (2017) 1147–1169

Nomenclature bi ( xi ) Control Gain

yi Plant Output
Vs Boost Input Voltage ui ∈ R The Control Input
Is Boost Input Current Pi ( x , x mi ) The Outside Disturbances
Vo Boost Output Voltage ei Ith Variable's Error Vector
Io Boost Output Current yd i Desired Output Trajectory's Vector
Db1 Disturbances Affecting The Boost Current yi Actual Output Trajectory's Vector
Db2 Disturbances Affecting The Boost Output Voltage M Number Of Rules
L Inductor ∼j
Fi Antecedent Membership Functions Of Xi
RL Inductor Resistance θj Centroids Of The Output Membership Functions
C Output Capacitor ∼j
μ Fj͠i Membership Grade Of Xi Related To Fi
RC Output Resistive Load Y Fuzzy Output
mi The System Degree P(N) Nth Sampling Time's Output Power Of The P(I) Curve
n The Number of Inputs I (N) Nth Sampling Time's Output Current Of The P(I) Curve
Xi The Ith Input State Variable D Duty Cycle
x i mi Mith Degree Time Derivative of The Ith Variable S Solar Irradiance Level
Ai ( xi ) Continuous State Function

more precisely, seem to be intelligent. This approach became very unknown mathematical model and the ability to estimate various
popular with the development of expert systems about thirty years ago. nonlinear functions are among the advantages leading to widespread
AI may be used to provide innovative ways of solving design issues and utilization of FL in various fields [7,9,29]. The fuzzy logic has also been
will allow designers to get an almost instantaneous expert opinion on widely utilized in the field of optimizing the power consumption of
the effect of a proposed change in a design [10,11]. In this regard solar arrays [30–33]. In this regard, compared to other AI methods
various AI methods such as Neural Networks (NN) [12], Fuzzy Logic such as NN, in case of proper tuning, the FL scheme has proven to
(FL) [13] and Genetic Algorithm (GA) [14,15] have been successfully exhibit superior performance in terms of efficiency and dynamic
implemented in various fields. Among the mentioned schemes, neural response [15].
based and fuzzy logic based approaches can be considered as the most Despite promising results presented in various researches [7,9,26–
common AI approaches in the field of MPPT [15,16]. 33], dependency of the FL performance to the expert's knowledge could
Neural networks have been implemented in various MPPT studies be counted as the main drawback of this approach. As a result, in the
[17–19]. In case of a comprehensive and extensive training routine, conventional type of this controller where the fuzzy parameters are not
based on a large set of data, which is generally time-consuming and properly tuned, mainly due to the forms utilized in designing the
complex to process, if the plant under control remains unchanged, the membership functions and the rule-base, performance deviations from
implemented NN method will perform satisfactorily without prior optimal expectations in environments with uncertainties are observed
knowledge of plant parameters [9]. [34].
Despite acceptable implementations, NN applications are bounded To overcome the effect of environmental conditions on controller
by its structural limitations. These limitations are mainly caused by the performance, minimize the need for designers experience, reduce the
nature of the pre-implementation approach utilized to train the NN. In rule-base size and improve the performance of fuzzy controllers, some
this regard, the following drawbacks have been pointed out in various efforts were made to tune the membership functions of the Fuzzy Logic
researches: Control (FLC) scheme [29,35]. In some researches, offline tuning of the
Using a very complex and time consuming training routine, usually membership functions were suggested. These offline routines consisted
prior to controller implementation, the NNs are trained to emulate the of heavy algorithms such as neural networks or even genetic-based
behavior of the specific under control plant. This is generally achieved algorithms which required a notable time taking training regime to
by utilization of a sufficiently large set of historical input-output data. perform suitably [36]. Using the offline processes to tune some of the
To ensure reliable operation of the NNs in a specific environment, the parameters made these routines unsuitable for real-time applications
relation between the nodes and different layers (the applied weights) and limited their implementations only to theoretical cases [37].
must be carefully and accurately determined by the applied training Moreover, it could be noted that in most of these documents,
routine. Hence, the PV module is tested for a long period (months or unrealistic assumptions such as considering the control gain, denoted
even years) so that the suitable input and the output relation of the by “bi”, as a known function [37], or as an undefined function with
NNs are obtained [7,9]. In this regard, the time-consuming nature of known sign [38] or even as an unknown constant [39], will also limit
the training process is one of the main drawbacks of the neural-based the applicability of the fuzzy logic to theoretical studies [40]. On the
schemes [20,21]. other hand, in most of the online adaptive schemes only one set of
Moreover, as mentioned in [22–25], neural networks are still membership functions (i.e. consequents) are tuned. In this regard, it
considered as unstable learning model regarding to the presence of will be shown that while tuning the antecedent and consequent
noisy sets, large training set, underfitting and overfitting problems that membership functions of a fuzzy- based scheme improves the overall
causes a lack of generalization and trapping on local minima solutions. system dynamics and steady states behavior, limiting the tuning
These drawbacks make the NN a less attractive approach of control process only to one set of membership functions (i.e. antecedent or
in practical and real-time implementations in environments where consequent membership functions), limits the controller's capabilities.
uncertainties and disturbances are constantly changing the model of Therefore, schemes which tuned all the membership functions online,
the plant and input-output relations under control. were focused on by researchers [40–42].
The FLC on the other hand has presented promising results under Moreover, many of the online methods tune the membership
varying conditions and perturbful, nonlinear and complex environ- functions based on heavy algorithms, increasing the overall system's
ments where the unknown plant model experiences variations through calculation burden [40–42]. Implementing such routines requires
time [26–28]. Simple linguistic terms, simple structure, easy imple- stronger ICs, where in such cases, the need for a stronger IC increases
mentation, capability to regulate complex nonlinear plants with the overall system cost. In this field of study, it can easily be seen that

1148
M. Nabipour et al. Renewable and Sustainable Energy Reviews 74 (2017) 1147–1169

no serious research regarding tuning approaches to tune the ante- [39], [37] and [38], “b” is considered to be a “time varying”, “unknown
cedent and consequent membership functions of an indirect fuzzy function” with “unknown sign and magnitude” respectively.
controller, online, using a simple algorithm without using unrealistic The proposed tuning routine adapts the MPPT control loops to the
assumptions has been reported. present environmental conditions. It also synchronously tunes the
Due to the importance of choosing a suitable MPPT algorithm, a antecedent and consequent membership functions online using a
review on available MPPT schemes is presented as the first contribu- simple algorithm, where it will be shown in the simulations section
tion in this paper. In the majority of the reviews dedicated to this field that tuning both the antecedent and consequent membership functions
of operation, the authors tend to examine the performance of various improves the performance of the controllers notably. The resultant
controllers with different natures. After presenting characteristics of controller is a robust, computationally light, online tuned FLC-based
these controllers, the superior controller is determined. For example, in scheme where all the membership functions are tuned synchronously.
[7–9], different classic MPPT approaches are compared with AI MPPT To exhibit the applicability of the proposed controller, simulation
schemes. Moreover, in [15], various AI methods are compared in terms results using the MATLAB-SIMULINK environment are presented. In
of various control indexes. Yet, no considerable review has been the field of MPPT and current reference tracking, the proposed method
conducted in which, apart from efforts to compare different classical is compared with conventional direct and adaptive indirect fuzzy
and AI approaches, the performance of sub-branches corresponding to controllers [37] to see the best tracking performance in following the
adaptation routines in MPPT controls and fulfilling design objectives desired power and current trajectory. The results exhibit the super-
are examined. Moreover, comparing different forms of FLC consisting iority of the proposed controller over conventional fuzzy and linear
of direct classic FL and adaptive fuzzy, especially in terms of adaptive controllers in terms of dynamic and steady states performances.
(online or even offline) characteristics in tracking the MPP has never This paper is described as follows. In Section 1, the introduction is
been performed before. In this regard, examining the available MPPT given and the main problem dealt with in this paper is described.
control methods throughout the presented review to find the optimum Where in Section 2 the literature review regarding the topic of MPPT
control approach in terms of capabilities to fulfill the defined MPPT control is presented. The literature review is then followed by PV setup
objectives, and proposing the optimum control routine characteristics and modeling in Section 3. The problem formulation and the proposed
has been considered as the main contribution and objective in this approaches are respectively given in Sections 4 and 5; where in Section
paper. On this basis, the efforts were put in obtaining the optimum 6 the simulation results are presented. Finally in Section 7 the
MPPT control characteristics through answering the following ques- conclusion is given.
tions throughout the review process:

1. Are the common classical MPPT routines, despite simple implemen- 2. Previous MPPT control works
tations and low cost, good choices to track the MPP desirably in
terms of dynamic speed, steady states smoothness and efficiency? In utilizing the solar energy, various factors are effective in
2. Despite high accuracy, do all the AI methods exhibit desirable MPPT optimizing the efficiency of PV modules. The efficiency of the structure
performance in terms of various control indexes? of the PV module (9–17%) which are compared in Table 1, operational
3. Is the classic fuzzy approach, with simple implementation charac- efficiency of the utilized inverter (95–98%) and the efficiency of the
teristics, an optimum approach to track the MPP at all instances? control structure to extract power from the PV module (higher than
4. Can impeccable MPPT performance be expected from FL control 98%) are among the most important factors [43,44]. In this regard, due
with a slight increase in the computational load caused by tuning to impossibility of changing the structure of the PV after installations,
only the consequent membership functions? modifying the MPPT controller is considered as the simplest method to
5. Can optimum MPP tracking behavior with computationally light improve the performance of the PV module. On this basis, due to the
characteristics be expected from available membership functions importance of the quality of MPPT control, objective control require-
tuning routines which tune the antecedent and consequent member- ments are expressed as fast dynamics, smooth steady states perfor-
ship functions online? mance, high efficiency, high reliability logical calculation complexity
and design cost. In this regard, different MPPT approaches were
On this basis, a review in three levels is presented in this paper. In presented and their ability to fulfill expressed control requirements
the first level, conventional MPPT algorithms, consisting of classic and was examined.in literature to improve the performance of PV setups.
soft computing MPPT methods are compared. After expressing general These approaches can be categorized in three main branches such as
characteristics of various conventional methods, AI approaches with offline, online and hybrid [45].
the potential to fulfill determined design objectives were selected as Offline MPPT approaches consist of methods where generating the
superior methods compared to classical MPPT methods. Then, by proper control signal to track the MPP is only achieved through
presenting characteristics of different AI methods, advantages and utilization of structural variables such as PV cell's short circuit current,
drawbacks of each scheme is determined; as a result, the FL was chosen temperature, irradiance level and open circuit voltage [47–49]. In this
as the advantageous and reliable MPPT approach among the AI regard, low accuracy and efficiency are considered as the mutual
methods. In the third level of the review, due to incapability of the drawback of such approaches.
classical fuzzy logic in fulfilling design goals at the same time and On the other hand, the online methods, without the need for any
improvement of the performance of the FL in case of desirable specific offline training procedures attempt to generate the desired
parameter tunings, various FL tuning and adaptation methods are control signal based on real-time PV output parameters. P & O [50], IC
examined and presented. After examining the characteristics of various [51] HC [52] are assumed as the most important classical online MPPT
adaptation methods, by focusing on online adaptation methods, the methods.
drawbacks of these methods are pointed out and because of the
Table 1
complexity and costly nature the reported tuning approaches where
PV structure dependent efficiencies [46].
all the membership functions are simultaneously adapted, the necessity
to improve the adaptation process is expressed. c-Si based Thin Film
To minimize the mentioned drawbacks, a novel MPPT controller is
presented as the second contribution of this paper. This MPPT Sc-Si mc-Si a-Si,Si/μc-Si CdTe CIS/CIGS

controller, consists of a fuzzy-based MPPT scheme tuned using a 14–20% 13–15% 6–9% 9–11% 10–12%
proposed novel membership function tuning routine; where unlike

1149
M. Nabipour et al. Renewable and Sustainable Energy Reviews 74 (2017) 1147–1169

In [50], Kota has utilized P & O method to track the MPP. In this modules connected to a high voltage BUS. To reach the optimum
research, they have initially examined the performance of the classical operation point in terms of the power extracted from the whole setup,
P & O method in tracking the MPP and have determined the advantages the MPPT problem was treated as a multi-objective optimization
and drawbacks of this type of MPPT control. In this research, it has problem. In this regard, the mentioned problem was optimized by
been said that despite easy implementation, the classical P & O method the PSO considering the relations between the output components of
presents undesirable behavior in MPPT tasks. In this regard, steady the PV module such as the PV current and voltage.
states power oscillations, undesirable dynamic performance and low In [56] Yang et al. Have benefited from the capability of the GA in
efficiency could be counted as the shortcomings of such MPPT finding a global optimum point to tune the structure of a hybrid solar-
methods. To overcome such characteristics, in [50] a tangent based wind setup. In this research, the authors have used the GA to increase
P & O routine is proposed. Based on the performed simulations, the systems reliability and optimize the performance of the system at
improvements in the performance of the MPPT scheme are confirmed. cases where the energy source is lost. In this regard, in the optimum
IC is the other classical MPPT controller which was examined in design procedure, various objectives such as the systems reliability in
[51] and implemented on a PV setup. In this research, similar to [50], all instances and the annual installation cost are considered. To solve
the authors initially attempt to examine the characteristics of conven- such a problem, variables such as the number and direction of the solar
tional IC control schemes in tracking the MPP. After determining the cell, the number of batteries and the wind turbines and the height of
drawbacks, similar to the ones exhibited by the P & O scheme, the the wind turbines were optimized by the GA.
constant step size is recognized to be the main cause of these draw- Also in [57], the GA was again utilized to optimize the performance
backs. Hence, in the presented solution, variable step size is utilized of a base controller in tracking the MPP. The utilized base controller is
instead of a fix step size. This modification in the nature of the step size a FL controller applied to a PV module feeding a simple load. In this
results in considerable improvements in the MPP tracking performance work, after comparing the performance of the conventional FL
compared to the classical MPPT routines. In this regard, increase in the controller with classical P & O MPPT controller, the stages of optimiz-
efficiency, diminishing the steady states ripples and improvements in ing the fuzzy controller were discussed. Following the optimization, the
the dynamic response are observed as a result of this controller resultant GA-optimized FL scheme was compared with other discussed
modification. This fact is confirmed in [51] using the performed schemes in tracking the MPP during various environmental conditions.
simulation and the practical implementation results. The results relate to the superiority of the genetic-based optimized FL
As the last classical MPPT algorithm, an example of successful HC scheme over conventional and traditional routine.
implementation is given in [52] by Loukriz et al. in this research, the As the last example of successful implementation of GA in MPPT
authors initially attempt to obtain a model for the PV cell. Following tasks, the research presented by Daraban et al. in [14] was examined.
the PV model determinations, considering general classical MPPT In [14], the GA is utilized to optimize the P & O algorithm in tracking
method drawbacks, a new converter referred to as the SEPIC converter the global MPP at conditions where multiple local MPP exist. By using
is presented. By implementing the HC algorithm on the PV cell using the P & O inside the genetic algorithm, the authors have managed to
the mentioned converter the MPP is properly tracked. The desirable decrease the calculation load of the optimization process along with
performance of the HC routine in combination with the proposed improving the performance of the P & O in tracking the global MPP.
converter is conformed using the simulations results. Independency to initial settings and capability to be implemented on a
The third branch in MPPT control is hybrid approach which is wide range of PV modules is considered as the characteristics of the
constructed by combining characteristics of the online and offline proposed scheme.
MPPT approaches. In the category, by combining a usually offline pre- FL on the other hand has also been successfully implemented to
implementation training routine and online process of real-time data, track the MPP. FL implementation in case of proper tuning presents
the suitable control signal to track the MPP is generated. This type of fast dynamics, moth steady states behavior and independency to plant
MPPT approach generally consists of optimization algorithms such as model and high robustness and efficiency. In this regard, in [58] Panda
PSO [52–55] and GA [14,56,57] and artificial intelligence methods et al. have attempted to implement the classic form of the FL to track
such as FL [37,57–61], NN [62–65]. the MPP of a specific PV setup. In this work, the fuzzy rules are
Among the mentioned AI methods, only the FL and the NN can be constructed based on operational principles of the classic P & O
directly implemented, while GA and PSO algorithms are usually routine. Based on the output variables of the PV module such as the
utilized to tune other processes. In this regard, some applications of output current, voltage and the extracted active power, the fuzzy output
the optimization algorithms are presented in [52–57]. is constructed and is directly fed to the DC-DC converter as the desired
In [53] Miyatake et al. have benefited from the PSO algorithm to duty cycle to track the MPP. In fact utilizing the FL to directly generate
control multiple-PV modules. The authors by confirming the complex- the control input is referred to as the direct fuzzy approach which will
ity of controlling multiple-PV modules using classical MPPT techni- be described later in this section.
ques, have proposed a novel PSO-based algorithm. In this algorithm, At the end, by comparing the MPPT performance of the conven-
using a multi objective function and by minimizing the installed tional direct fuzzy with the classical P & O approach, the superiority of
sensors the MPPT of the PV module is successfully controlled in this routine with respect to various control indexes is demonstrated.
various operational conditions. Similar studies were also conducted in [59] where equivalent results
In [54], the PSO routine is utilized to improve the MPPT control were obtained.
performance by Shaque et al. in this research, the main MPPT control Another application of the indirect fuzzy MPPT control is demon-
routine is selected to be HC and hence, the PSO is utilized to optimize strated in [60] by Chiu et al. In this study two new solutions are
the performance of the classical HC algorithm and directly generate the proposed as MPPT controllers. Both of the proposed solutions are
required converter duty cycle. Following this optimization, the resul- simultaneously implemented and compared. The first solution is
tant HC algorithm is then compared with conventional classic MPPT referred to as the maximum power-voltage based routine. In this
algorithms where the improvements are clearly observed. routine, after receiving the PV current, voltage and power signals, the
In [55] Renaudineau et al. by refereeing to the capabilities of the reference PV output voltage corresponding to the MPP is generated.
PSO algorithm compared to other optimization routines corresponding The generated reference voltage is then fed to the FL controller in
to simplicity of implementation, higher convergence speed and in- charge of generation of the duty cycle. In this case, the objective of the
dependency to the initialization point, have utilized the PSO algorithm FL controller is to minimize the error between the PV output voltage
to improve the MPPT performance. In this regard, the MPPT task was and the generated reference voltage.
considered for a PV setup consisting of multiple series connected PV In the second solution, as the output PV signals are received, the

1150
M. Nabipour et al. Renewable and Sustainable Energy Reviews 74 (2017) 1147–1169

controller attempts to minimize the slope of the P-V diagram by environmental parameters, a reference voltage to maximize the PV
varying the output voltage of the PV module. It could easily be output power is derived. This reference voltage is then fed to a linear
confirmed that as the operation point approaches the MPP, the duty cycle controller, where the objective of the controller is to
magnitude of the P-V slope decreases; where in the case where the minimize the error magnitude between the reference and the actual
operation point is located directly on the MPP, the magnitude of the PV output voltages. Finally, the authors have confirmed that by using
slope is minimized to zero. To evaluate such MPPT solutions, practical such solutions, the number of sensors required for practical imple-
implementations have been performed, where the resultant improved mentation is reduced and the MPPT efficiency is improved.
MPP tracking performance is clearly observed. Following the research presented in [62] Ocran et al. have
Another application of fuzzy control is presented in [61] by El presented a different method of PV modeling and MPP tracking by
Khateb et al. in this study the FLC is used to generate the duty cycles of utilizing NN [63]. In this study, with a DC motor as the load, similar to
the combination of a SEPIC converter and an AC inverter. On this the approach utilized in [62], initially an accurate model of the PV is
basis, in this study two individual FL controllers are utilized. Tracking obtained by training the NN using a sufficiently large set of training
the MPP by proper DC converter duty cycle generation is defined as the data. On this basis, for each combination of the environmental
main objective of the first FL controller, while in the second FL temperature and irradiance, the current and voltage corresponding to
controller, generation of a suitable load voltage is considered as the the related MPP are obtained. Using the obtained current and voltages,
main objective. In this regard, by sensing PV output parameters, the the authors attempted to calculate the corresponding maximum power.
reference output voltage to track the MPP is generated. This reference The resultant maximum possible power is then fed to the linear duty
voltage is then fed to the first FL controller and the suitable control cycle regulator, where the efforts are put into minimizing the error
output to regulate the SEPIC converter is generated. The SEPIC's magnitude between the actual and the reference output powers
output voltage feeds the DC-link coupled to the inverter, where by generated by the NN scheme. In similar research conducted in [64],
proper inverter control, the demanded AC voltage is generated. As a similar performance was expected from the NN.
result of such systems, apart from improvements in the MPP perfor- following the researches performed to implement NN, a novel
mances, the THD of the AC system is also reduced. approach is presented in [65]. In this study, instead of utilizing the
In the other form of implementation of the FL, referred to as the NN to model the plant as a whole and to determine the optimum
indirect fuzzy approach, the control input is not directly generated by voltage or power, a multi-layer NN is proposed to estimate nonlinear
the FL scheme. In fact, in this type of application, the FL is used to functions of the dynamic equations. In this regard, by accurately
tune, adapt and improve the performance of another controller. In this estimating the state functions using offline training data, state-feed-
regard, in [37] this branch of applications of FL is examined. In this back based schemes are utilized to minimize state errors. Considering
study, the MPP is tracked using two levels of control loops. In the first the mentioned discussions, characteristics of various AI MPPT meth-
loop, the physical position of the PV is determined in a way so that the ods are given in Tables 2–4.
maximum amount of irradiance possible is received. Following the In various researches, comparisons between the classical and AI
optimization of the PV position, a fuzzy adaptive P & O routine is MPPT algorithms are performed. In these reviews, drawbacks such as
utilized to track the MPP. In this type of application, the FL tunes the P slow dynamics, high steady states oscillations and low efficiency were
& O parameters online based on real-time PV output parameter data. reported as classical MPPT control characteristics. Hence, it is easily
On this basis, the desired PV output voltage corresponding to the MPP concluded that the classical MPPT methods fail to fulfill the control
is generated by the P & O algorithm. The P & O output is hence directly requirements and therefore cannot be considered as the optimum
fed to a linear PI controller, acting as the duty cycle regulator, so that control approach. On the other hand, advantages of AI methods in
the desired duty cycle to track the MPP is generated. Having the terms of higher efficiency, lower steady states oscillations, higher
desired duty cycle generated by the PI controller, it is applied to the tracking speed and reliability are confirmed. In this regard to examine
DC-DC converter using simple PWM methods. At the end, the validity the performance of different MPPT schemes, in [8] by Mellit et al., a
of this approach is confirmed using simulations and practical imple- comprehensive review and comparison of different MPPT approaches
mentations. is presented. In this review, after listing various MPPT approaches, the
NN is also an attractive approach of control in which a model of a authors have presented a comprehensive categorization based on pre-
specific plant is generated through a comprehensive training process. implementation and during implementation requirements. On this
In [62], a NN scheme is utilized to track the MPP corresponding to a basis, the MPPT schemes are categorized into offline, online and hybrid
PV module installed to feed a solar vehicle. In this regard, similar to the schemes. Following this categorization, from each branch a control
fuzzy approach presented in [60], output voltage generation consists of method is selected and compared with other present MPPT routines in
two stages. Initially, by training the NN offline, using an adequately terms of dynamic speed, efficiency and implementation simplicity.
large set of experimental PV input-output data, the relation between Where as a result of this comparison, superiority of the AI methods in
the environmental conditions and the reference output voltage corre- terms of dynamic speed, tracking efficiency and steady states oscilla-
sponding to the MPP is obtained. As a result, for every combination of tions over conventional classic MPPT schemes is proven.

Table 2
Optimization Based MPPT Routines.

Authors Year Application

PSO Based Optimization MIYATAKE et al. 2008 Search for global MPP with reduced number of sensors
Kashif Ishaque et al. 2012 Tune HC-Control MPPT using a Buck-Boost Converter
Letting et al. 2012 Act as a C-Mex S-function to tune FLC rules
Renaudineau et al. 2014 Optimization of operation point of multiple-PV modules connected in series attached to a high voltage BUS to
achieve MPPT

Genetic Based Optimization Yang et al. 2008 Optimal structure defining of a hybrid wind-solar setup
Larbes et al. 2009 Tune the FLC to improve MPPT performance
Daraban et al. 2014 Tune the P & O to improve MPPT performance
Lasheen et al. 2016 Tune the PI controller to improve MPPT performance

1151
M. Nabipour et al. Renewable and Sustainable Energy Reviews 74 (2017) 1147–1169

Table 3
Classical Fuzzy Logic based MPPT control.

Authors Year Control Variable Converter Type Controller Type Application

Panda et al. 2011 DC-Converter Duty Cycle Buck-Boost Direct The desired Voltage is generated by P & O and fed to the FL by which the
duty cycle is generated
Chiu et al. 2011 Duty Cycle Buck Direct FL is implemented for MPV based routine and direct maximum power
control.
Khateb et al. 2012 DC-Converter Duty Cycle and Sepic and AC Direct Double FL is utilized to generate duty cycle of the SEPIC converter with the
AC inverter switching signals inverter reference voltage being generated by the P & O algorithm. An AC inverter is
also switched using the second converter.
Nabulsi et al. 2012 P & O parameters Buck Indirect FL is used to tune P & O in tracking the MPP
Bendib et al. 2014 DC-Converter Duty Cycle Buck-Boost Direct The desired Voltage is generated by P & O and fed to the FL by which the
duty cycle is generated

Following the research given in [8], another comprehensive com- been presented in [15,16]. In these researches, the FL is assumed to be
parison of various MPPT schemes is given in [46]. Similar to the review properly tuned and the NN is also considered to be well trained using
given in [8], routine categorization is also performed in [46], where an adequately large set of data.
based on experimental results, superiority of the AI methods over In [15], the FL and NN controllers are applied to track the MPP
classical MPPT schemes is confirmed. Moreover, the authors have also throughout the experimental validations. The utilized PV module
listed the future work expectations on renewable energy sources. In this consists of two parallel Siemens SM50-H 50w arrays feeding a 24 V
regard, by grouping the different aspects and requirements for load. In this research, the comparison between the FL and the NN is
implementation of the MPPT schemes to branches such as control based on practical evaluations. In this regard, the FL and the NN are
module, Fault-detection and diagnosis module, Monitoring and super- respectively tuned and trained using data gathered from the operating
vision module and Prediction module, application of AI in improving system. Based on the experimental results, the authors have proven
the operation of each module is demonstrated. that by applying the FL, the MPP is followed with lower error, higher
In [66] a thorough examination is presented and different MPPT accuracy and increased output power compared to the NNs.
schemes are analyzed from different aspects. After expressing general In the review given by Chekired et al. in [16], various FL and NN
characteristics of different approaches, the authors have divided the schemes such as randomly tuned FL, desirably offline tuned FL, ANN
MPPT schemes to two branches of classical and soft computing MPPT and ANFIS were compared. These schemes were subjected to various
methods. On this basis, all the sub-branches of each category are then environmental variations such as constant variations in irradiance and
compared through pre-determined criteria. At the end, by presenting temperature, rapid variations of irradiance and rapid variations of
the advantages and disadvantages of each scheme, the authors have temperature. In all the mentioned cases, the desirably tuned FL
confirmed the superiority of the soft computing or AI methods over exhibited superior performance compared to other AI schemes includ-
conventional classic MPPT techniques. ing the ANN. This performance improvement has been evaluated using
In this regard, a comparative study on both AI and conventional various control indexes such as higher efficiency, higher response speed
MPPT methods has been performed by Mellit and Kalogirou in [67]. In and lower steady states oscillations compared to the NN method. The
this research, various MPPT techniques were implemented, tested and results of these simulations were also confirmed by experimental
compared under different conditions. In this review, based on various results. Hence, judging based on the mentioned NN and fuzzy
simulation results, the authors have confirmed that by utilizing AI characteristics as their drawbacks and merits, and the conclusions
methods, the dynamic and convergence speeds increase, control given in [15,16], the FL can be considered as a reliable AI choice in
performance and efficiency are improved and steady states oscillations tracking the MPP. The results of these comparisons are also presented
are decreased. Moreover, possible divergences in variable weather in Tables 7 and 8 [15,16,69].
conditions are minimized. Similar results have been reported in other Despite desirable dynamics and steady states behavior and high
researches such as [7,9]. Moreover, considering the above character- efficiency when properly trained, due to the complex training routine
istics, a brief comparison over the general MPPT schemes is given in required by the NN approach, parameter dependence of this approach
Table 5. and impossibility to generalize a pre-tuned NN to control multiple
Corresponding to the comparisons made in Tables 5 and 6 and environments, based on findings given in [69], the NN approach
reference [7,68], the AI routines are considered as the superior cannot be considered as a reliable MPPT method. Moreover, the overall
approaches. In this regard, due to optimizing nature of GA and PSO cost of designing and implementing an NN scheme due to the large set
algorithms, only FL and NN methods can be considered as direct of data and complex algorithms is generally high. Hence, the NN fails
control approaches. To determine the suitable AI approach in fulfilling to fulfill all the control requirements at the same time. On the other
control requirements, numerous researches have been conducted based hand, Since the NN is trained to control a specific plant in a particular
on comparison of these two approaches. environment, it is considered to be a heavily environmental dependent
To ease the process of choosing between the NN and FL to perform solution. In this regard, when implemented as an MPPT controller, in
MPPT tasks, extensive comparison between the FL and the NN have case of any changes in the structure of the PV module, the training

Table 4
NN-based MPPT control.

Authors Year Control Variable Converter Type Modeled Plant Application

OCRAN et al. 2005 MPP Voltage Boost PV-Module The PV structure is modeled by NN and the reference MPP voltage is generated
Bahgat et al. 2005 MPP Power Buck PV-Module The PV structure is modeled by NN and the reference MPP Power is generated
Ramaprabha1 et al. 2009 MPP Power Boost PV-Module The PV structure is modeled by NN and the reference MPP Power is generated
Lin et al. 2011 MPP Voltage Boost PV-Module The PV structure is modeled by NN and the reference MPP Power is generated
Kassem 2012 State Functions Buck-Boost State Functions Multi-level NN is utilized to estimate state functions used in feedback linearization
approaches

1152
M. Nabipour et al. Renewable and Sustainable Energy Reviews 74 (2017) 1147–1169

Table 5
AI and classical MPPT control characteristics [68].

Classical MPPT Control AI MPPT Control

Method P&O IC HC GA PSO NN Classic FL

Advantages • Simple • Highly efficient • Easy construction • wide search • Effective usage of • Effective for all • Highly robust
construction for steady • Most proficient possible search space types of • Accurate error
• Easy change in algorithm in slow • can be • Highly effective in irradiation detection
implementation irradiance irradiance changes applicable to search for global conditions • Very effective
• Less sensor • Simple and fast changes in maximum • Accuracy in when combined
requirements highly reliable irradiance • Simple and reliable solutions with other AI
• Highly
handling
effective in methods

nonlinearities

Method P&O IC HC GA PSO NN Classic FL

Disadvantages • High
perturbation
• Power
frequency
• Less capable
handling
in • High memory • Initializations
requirements parameters are
of • High computational • Hard
time converge with
to

rate oscillations dynamic changes • Large difficult • High memory fast irradiance
• High • Poor in irradiance computation • Computation is requirements variations
oscillations convergence • Useful only in low process difficult in large • Heavy training • Rules cannot be
around MPP power • Time consuming populations required changed once
applications training routines • High plant and
Environment
defined

dependence
• Low reliability
when uncertainties
are present

procedures are to be repeated. On the other hand, it has to be noted time and hence is considered as the main control logic to track the MPP
that the structure of PV arrays are time-variant. This means that the in this paper.
structure of a specific PV module experiences various changes over Despite all the mentioned advantages and constructive character-
time. As a result, to ensure reliable performance of neural based istics, shortcomings such as dependency to expert knowledge have
schemes, periodical training procedures have to be considered, where limited the application of fuzzy based MPPT schemes. To minimize this
in each training procedure a new set of significantly large set of training drawback, efforts have been made to tune the FL parameters [70–84].
data is processed by a complex optimization algorithm. Besides In the initial form of these efforts, pre-implementation offline tuning
variations in the structure of the PV module, some other common using optimization algorithms such as the GA [70–76] and the PSO
phenomena such as unforeseen weather conditions, unpredicted [77–85] are considered.
climate changes or even changes in the converter circuit caused by Tuning FL parameters using GA has been reported in various
temperature variations force the training routine to be repeated again. researches. For example, in [70], Messai has managed to optimize the
It should also be noted that for real-time implementation of a neural- fuzzy membership functions using GA. It is shown that by successful
based scheme in practical environments where the system dynamics GA implementation and proper tuning of fuzzy parameters, the
are constantly subjected to parameter variations, the number of performance of the genetically tune FL exhibits notable improvements
periodic training procedures increase significantly to ensure reliable compared to the conventional FL controller. Apart from MPPT GA-FL
operation of the control scheme. Due to the computational complexity has seen much attention in various fields [71–76]. In this regard, data
and time-consuming nature of the training process, for the scheme to mining [71,72], pixel classifications [73,74], time series forecast [75],
be practically implementable for real-time applications, much stronger mining fuzzy association rules [76], and etc… could be pointed out as
micro controllers with higher computation power must be utilized. GA-FL reported applications.
Such micro controllers significantly increase the overall cost of the Apart from the GA, the PSO algorithm has also seen much attention
controller design [7,9,22–25]. in the field of fuzzy parameter optimizations in MPPT applications
Moreover, based on the results given in [69] corresponding to the [77–81]. Other than in the MPPT control field, PSO-based fuzzy
superiority of the well-tuned FLC over fully trained NN approaches in optimization has also been applied in the fields of robotics [82], grid
terms of tracking error, efficiency and output power, this logic is found frequency control [83], identification on smart phones [84], load
to present potential of fulfilling all the control requirements at the same frequency control [85] and etc….

Table 6
AI and classical MPPT control characteristics.

Classical MPPT Control AI MPPT Control

Method P&O IC HC GA PSO NN Classic FL

Tracking Speed Low Low Low High High High High

Convergence Speed Low Low Low High High High High
Complexity Low Low Low High High High Low
Implementation Easy Easy Easy Difficult Moderate Moderate Moderate
PV Dependent No No NA No No Yes No
Prior Training No No No Yes Yes Yes Yes
Efficiency Low Moderate Moderate High High High High

1153
M. Nabipour et al. Renewable and Sustainable Energy Reviews 74 (2017) 1147–1169

Table 7
Comparison of Fuzzy-based and Neural based schemes.

Controller Complexity Response Time Efficiency Required Sensors Steady States Oscillations Overall Cost

Optimally Tuned FL Complex Very fast Most efficient Current, Voltage Negligible Medium
Classic FL (No Tuning) Simple Fast Efficient Current, Voltage Low Medium
Optimally Trained NN Complex Fast Efficient Voltage, Temperature, Irradiance Low High
ANFIS Complex Fast Efficient Voltage, Temperature, Irradiance Low High

Table 8 cases desirable operations were observed and reported. In another

Comparison of Fuzzy-based and Neural based schemes During Various Tests. attempt by Barkat et al., estimation of state functions using adaptive
consequent fuzzy membership functions has been implemented in [37].
Controller Fast Temperature variations Fast Irradiance variations
Robust control of the PMSM drive using desirable fuzzy-based state
Efficiency (%) Response Efficiency (%) Response function estimations in tracking a reference speed trajectory has been
time (s) time (s) presented as the main control objective in [37].
Moreover, successful control of the inverse pendulum and magnetic
Optimally 98.2 2.97 97 3.12
Tuned FL
levitation using adaptive fuzzy-based function estimations has been
Classic FL (No 98.07 8.23 96.4 8.15 reported in [88]. The fuzzy logic has directly been utilized in this study
Tuning) where the output of the FL is adapted using simple Lyapunov's theory
Optimally 97.86 7.75 96.66 7.84 formulations. In all the implemented methods, Large calculation load
Trained NN
forced to the system by the tuning routine, time-consuming training
ANFIS 97.52 7.87 96.25 7.92
procedure and dependency to training data and most important of all,
the offline nature of the antecedent membership functions tuning
Despite desirable operations in predetermined conditions, this type procedures are considered as notable drawbacks of the “consequent
of tuning is considered to be an offline routine and will deviate from adaptive” adaptation routines. Moreover, it will be shown that by
desirable operational characteristics when implemented to control real- utilization of adaptive output membership functions and stationary
time applications in environments where disturbances, nonlinearities offline-tuned antecedent membership functions, the performance of
and perturbations exist [37]. Thus, this is considered as the obstacle in the adaptive fuzzy logic experiences some limitations.as a result of
the process of optimal performance of FL. As a result of these these limitations, the adaptive fuzzy controller will deviate from
deviations, besides time consuming implementation process, the offline desirable performance characteristics. And hence fails to fulfill these
tuned FL fails to fulfill control requirements and hence, cannot be control requirements. These limitations are not observed while adapt-
considered as the target optimum control method. ing the antecedent and consequent membership functions simulta-
To reduce the sensitivity of the fuzzy controller to disturbances, neously, as proposed in the “antecedent-consequent adaptive” routine.
nonlinearities and perturbations, efforts were made to adapt the As a result, the control performance is improved and the online
structure of this logic online [37]. In this regard, this type of logic adapted “antecedent-consequent” presented potentials to optimally
can be analyzed from two points of view. From the application point of fulfill the control requirements.
view, the adaptive fuzzy logic is categorize into two branches namely To minimize the time consumption, and dependency to the training
direct and indirect adaptive fuzzy approaches. The difference between data or even the designer's experience related to the pre-implementa-
these two approaches is mainly due to the routine utilized to generate tion training routine, various simultaneous adaptation routines were
the control signal. In the direct FL approach, the output of the FL proposed by researchers in [40,41,91–94]. In the approach presented
controller is directly applied to the plant as the control input. While in in [91], all the consequent and antecedent fuzzy membership functions
the indirect approach the FL is utilized to tune other control schemes are adapted and obtained online. In this research, by defining various
[38]. The most important application of the adaptive indirect fuzzy objective functions with different weighing factors, using complex
approach is in estimating various nonlinear functions. In this type of optimization algorithms, the antecedent membership functions are
application, the adaptive indirect fuzzy approach benefits from the simultaneously tune with the consequent membership functions as the
global estimation theory. Based on the global estimation principles, the environmental conditions and the operation point change. Following
fuzzy logic can be utilized to estimate any nonlinear but bounded the research presented in [91], in [41] simultaneous adaptation of
function on a compact set [86]. fuzzy membership functions is done by limiting the possible member-
Apart from the application point of view, the fuzzy logic can also be ship function forms to specific pre-defined shapes. In these shapes, for
categorized from the turning point of view to “consequent adaptive” each membership function three structural parameters are considered
and “antecedent-consequent adaptive” routines. The viewpoints of such as the center of the membership function, the left and the right
these two approaches coincide in the method used to tune the point of the membership function. For each membership function,
consequent membership functions. While the methods suggested to various possible combinations are constantly checked with an algo-
tune the antecedent membership functions are completely different. In rithm similar to PSO so that the optimum membership function shapes
the “consequent adaptive” approach the antecedent membership func- could be reached.
tions are tuned offline similar to methods used in [70–84] to tune the Another simultaneous tuning algorithm is presented in [40] by M.J.
classic direct fuzzy approach [37,87–90]. and Deng. in this work the simultaneous adaptation of the membership
Therefore, in [87] the fuzzy adaptive approach with adaptive functions is achieved through implementation of a objective function
consequent membership functions has been utilized to estimate various based Q-learning approach. Similar to [41], the implemented algorithm
nonlinear state functions and control input gain. In this study, the is of optimization search type where by which a large calculation
authors benefit from hybrid adaptation laws to guarantee accuracy of burden and a time-consuming heavy training procedure is forced to the
estimations. In this type of adaptation, the modeling and tracking control system.
errors are utilized where the control efforts are made to minimize these Yu et al. have also presented a novel membership functions tuning
errors. To evaluate the performance of the proposed scheme, it has routine in [94]. It is clearly observable that tuning the membership
been implemented on inverse pendulum and chaotic systems.in both functions is achieved using complex and heavy algorithms in this

1154
M. Nabipour et al. Renewable and Sustainable Energy Reviews 74 (2017) 1147–1169

report. Moreover, the authors have pointed out that in the adaptation nonlinear characteristics include the relations between the output
algorithm, the data are initially categorized using clustering methods. voltages (V), output current (I) and the Output Active Power (P). In
Following this categorization, using a heavy learning based approach this regard, to benefit the most from this source of energy, the
with variable rate is utilized to adapt all the membership functions and maximum power must be provided by the PV-cells, where the efficiency
modify fuzzy parameters. is maximized. The maximum power is generated at a point referred to
Gradient based approaches have also been reported among com- as the Maximum Power Point (MPP). The MPPs and the MPP
mon approaches utilized in various researches to tune and adapt the variations caused by changes in the environmental conditions are
antecedent and consequent fuzzy membership functions at the same respectively determined by black dots and dashed line in Fig. (1a). In
instants, online [92,93]. These methods, apart from exhibiting slow this figure, various solar irradiance levels are denoted by Si, where i=1,
convergence speed, force a heavy calculation load to the system and are 2, 3, where the corresponding MPP currents are also denoted as IMPPi.
generally only implementable to systems where obtaining gradients of In this regard, Fig. (1a) demonstrates gradual increase in irradiance
nonlinear functions is possible. levels denoted by Si and their effect on the MPP. In this figure the
High complexity of the adaptation algorithms, high computational relation between irradiance levels are defined as S1 < S2 < S3.
load and high design and implementation cost can be considered as the Moreover, the setup utilized to track these MPPs is also demon-
mutual drawback of these adaptation routines. It can easily be observed strated in Fig. (1b). This setup consists of a PV-module and a Boost-
that none of the mentioned schemes in this section manage to fulfill the converter to control the PV-module output power.
control requirements at the same time. The performance of the As depicted in Fig. (1b), a DC/DC converter (Boost-Converter) is
discussed methods in fulfilling control requirements can be listed as utilized to obtain a smooth power trajectory and track the MPP by
follows: controlling the PV-array output current and hence, output power.
Based on this figure, multiple steps have to be taken. Initially, the
• Classical MPPT routines fail due to low dynamic speed, high steady actual position of the Operation Point (OP) has to be determined.
states oscillations and low efficiency. According to Fig. (1a), since the relations between the current and
• The NN as an AI method also fails due to environmental dependent power of individual PV-arrays differ from one to another and even
and time-consuming training routine leading to low reliability. change from on environmental condition to the other, the OP cannot be
• GA and PSO cannot even be considered as direct control methods determined based on mathematical relations. Hence, an online algo-
since they can only be used to optimize other controllers and rithm is utilized to achieve this task. In this algorithm, the present
processes. output power and current of the PV-array are compared with the same
• The classical FL also fails since it requires proper tuning to show parameters sampled in the previous sampling time. Using these set of
desirable performance at all instances. data, the present location of the OP is estimated.
• Although “Consequent-adaptive” shows improved performance, but Having determined the present OP, a reference MPPT current
it also fails since it requires tuning the antecedent membership signal denoted as IMPP* is generated by the MPP generator. This
functions similar to the classical fuzzy approach. current, will be modified in a manner so that it tends towards IMPP,
• The present “Antecedent-Consequent” methods also fail due to very acting as the reference current for the Boost converter. As a result,
high training complexity leading to costly implementations. IMPP* has to be drawn from the PV-Module so that the MPP is
tracked. Considering the Boost converter positioned at the inner
To overcome these drawbacks and yield an MPPT control approach control loop, it must have much faster dynamics compared to the
capable of fulfilling all the control requirements at the same time, in outer loop components such as the MPP generator. With fast enough
this paper a novel tuning routine and as a result, a novel controller in Boost converter dynamics, MPP reference current (IMPP*) can easily be
which the antecedent and consequent membership functions are drawn from the PV-module. Therefore, without loss of generality, the
simultaneously adapted online using a computationally light algorithm transfer functions mapping the IMPP* to the Boost input current “Is”
I
is proposed. Using the proposed scheme the nonlinear functions, can be approximated as Gboost (s ) = I s * ≅ 1 . Hence, it is observed that
MPP
disturbances, nonlinearities and perturbations are estimated and by *
any variations in the IMPP results in variations of the PV-module's OP.
generating the suitable control signal, with the steady states and As a result, accurate modifications of the IMPP* can lead to desirable
dynamic performance improved, the stability of the system is guaran- MPPT behaviors where the OP is moved towards the MPPs depicted in
teed. Fig. (1a).
In this process, the IMPP* is generated using the flowchart given in
3. PV setup and MPPT modeling Fig. (2a). This flowchart is easily transferred to a fuzzy-based function
where it is used in the fuzzy-based MPP-generator. The membership
In this section, the PV-setup model is examined. The utilized model functions and the rule-base related to this fuzzy-based MPP-generator
is depicted in Fig. (1a) and described in [30]. The PV-cells, being are given in Fig. (2b) and Table 9 where the control principles and the
dependent on the environmental conditions, exhibit nonlinear char- control schematics are respectively depicted in Figs. (2c and d).
acteristics when subjected to ambient parameter variations. These In Fig. (2c), the variations of the IMPP* based on the present PV-

Fig. 1. PV-array output curves and operating setup, (a) PV Characteristics Curve (b) PV-Setup for MPPT.

1155
M. Nabipour et al. Renewable and Sustainable Energy Reviews 74 (2017) 1147–1169

Fig. 2. : IMPP* Generator (a) IMPP* Flowchart (b) IMPP* Generator Membership Functions (c) IMPP* and Adaptive Gain Principles, (d) Controller Schematics.

1156
M. Nabipour et al. Renewable and Sustainable Energy Reviews 74 (2017) 1147–1169

Table 9 system requirements.

Rule-base to generate the IPR signal.
xi mi = Ai ( xi ) + bi ( xi ) ui + Pi ( x , xi mi )i =1,..., n
ePI/δePI N Z P yi = xi (7)

N NB NM NS The main goals in designing the control input are to minimize the
Z PS Z NS tracking error and to guarantee plant's stability. In this regard,
P PS PM PB depending on the assumptions considered, the controller design can
be treated in two ways, the ideal and the realistic approaches. In this
section, the ideal approach is discussed while the realistic approach is
module's output is demonstrated, where:
treated in the next section.
ePI = [P (n ) − P (n − 1)][I (n ) − I (n − 1)] (1)
A. Ideal approach:
δePI = eP (n ) − eP (n − 1) (2)
The principles of the MPP generation can easily be understood In the Ideal approach, Ai ( xi ) and bi ( xi ) are considered to be
through Figs. (2b) and (c). As an example, for positive ePI and δePI , it is measurable and Pi ( x , xi mi ) is neglected. In this regard, considering
inferred that the OP is approaching “1” from “2”. Similarly for negative x = [x1, x2, ... ,xn]T ∈ R n , x i = [xi , xi̇ , ... ,xi (mi −1) ]T ∈ R mi and x = [ x1, x 2, ... , x n]T ∈ R n × mi ,
ePI and δePI , the OP is approaching “3” from “2”. Other cases are ei , yi and yd i are formulated as:
inferred in a similar manner. In fact, in all the possible variation in
yi = [ yi , yi̇ , ... ,yi (mi −1) ]T ∈ R mi (8)
ePI , δePI the present position of the OP is determined by the ePI while the
movement direction of OP is defined by δePI sign. Based on values of yd i = [ ydi , ydi̇ , ... ,ydi (mi −1) ]T ∈ R mi
ePI , δePI the required actions are taken by the IMPP* generator using (9)
the scheme depicted in Fig. (2d). In this regard, the fuzzy-based ei = yd i − yi = [ei , ei̇ , ... ,ei (mi −1) ]T ∈ R mi (10)
controller, with the membership functions and rules, as depicted and
presented in Fig. (2b) and Table 9 is utilized to control the IMPP* as: using (7)−(10), ei dynamics are obtained as:

IMPP* = ∫ Y (ePI , δePI ) T × dt ei̇ = Fi ei + Gi ei mi = Fi ei + Gi ( ydi (mi ) − Ai ( xi ) − bi ( xi ) ui − Pi ( x , x mi )

(3)
ei = CiT ei
To overcome the common drawback observed in conventional
(11)
direct fuzzy schemes, which is high frequency oscillations about the
MPP, the adaptive gain “T” dynamics present in Fig. (2c) and is defined where:
as follows: ⎡0 1 0 ⋯ ⋯ 0⎤
⎢0 0 1 0 ⋯ 0⎥ ⎡1⎤
T = T1 × tanh (T2 × |ePI |) (4) ⎢ ⎥ ⎡0⎤ ⎢⋮⎥
⋮ ⋮ 0 ⋱ ⋱ ⋮⎥
Fi = ⎢ Gi = ⎢ 0 ⎥ Ci = ⎢ ⎥
where T1 and T2 are constants. Using this gain, as the OP tends ⎢⋮ ⋮ ⋮ ⋱ ⋱ 0⎥ ⎣⎢ 1 ⎥⎦ ⎢0⎥
towards the MPP, variations of IMPP* will be limited. Thus, oscillations ⎢0 0 0 ⋯ 0 1⎥ ⎣0⎦
⎢⎣ 0 0 ⋯ ⋯ ⋯ ⎥
0⎦ (12)
around the OP will be minimized. These oscillations in the drawn
current result in severe ripples in the generated power by the PV- By considering vi as vi = ydi (mi ) + KiT ei , (5) is modified as:
module, which in fact, damage the electrical instruments. In the
simulations section, a comparison between the proposed schemes with ei̇ = Fi ei + Gi ei mi = Fi ei + Gi (vi − KiT ei − Ai ( xi ) − bi ( xi ) ui − Pi ( x , x mi )
the conventional direct fuzzy routine regarding the “active power” = (Fi − Gi KiT ) ei + Gi (vi − Ai ( xi ) − bi ( xi ) ui − Pi ( x , x mi ) (13)
oscillations is presented.
Having the reference MPP current IMPP* generated, it is fed to the ei = CiT ei (14)
Boost-Converter to be drawn from the PV-module. By accurate control
of the Boost-Converter, IMPP* will be tracked desirably and hence, where KiT = [Ki,1, Ki,2, ... ,Ki, mi ] ∈ R mi are selected in a manner so that
(Fi − Gi KiT ) is stable.
desirable MPPT is guaranteed. In this regard, the Boost regulator plays
In the ideal approach, considering the ideal problem assumptions,
a vital role in the MPPT procedure.
where Ai ( xi ) and Bi ( xi ) are considered to be measurable and Pi ( x , xi mi )
is neglected, applying the control output uideal as
4. Problem statement 1
uideal = b (vi − Ai ( xi )), the following is concluded:
i
1
As mentioned in the previous section, desirable MPPT operation is ei̇ = (Fi − Gi KiT ) ei + Gi (vi − Ai ( xi ) − bi ( xi )( b (vi − Fi ( xi )))
i
highly dependent on the BOOST current regulator. The proposed
= (Fi − Gi KiT ) ei (15)
control scheme is designed to regulate the BOOST current so that by
tracking IMPP* desirably, the MPPT task is achieved properly. Hence, hence:
in this section the general control problem dealt with in designing an
ei (mi ) + ki, mi ei (mi −1) + ... +ki,1 ei = 0 (16)
MPPT controller, where the proposed adaptive fuzzy-based scheme
used to control the Boost-Converter is discussed. In this regard, the Thus, according to (16), we will have lim ei (t ) = 0 , and hence, the
Boost control dynamics are given as: t →∞
asymptotical stability of the ideal system can easily be concluded [95].
1 Yet, in practice, since accurate plant model is not available, Ai ( xi ) and
Iṡ = (−R + Vs − Vo (1 − D )) + dB1
L (5) are generally immeasurable. Moreover, system interactions and dis-
turbances denoted as Pi ( x , xi mi ) have to be accounted for due to various
1
VȮ = ((1 − D ) Is − Io ) + dB2 disturbances and uncertainties, making the ideal problem assumptions
C (6)
invalid for realistic applications.
System dynamics given in (5) and (6) can clearly be treated as the To design a suitable control law, the plant model or the state-based
Multi-Input Multi-Output (MIMO) system dynamic equations given in functions have to be estimated and the disturbance terms,Pi ( x , xi mi ),
(7). Hence, in this research, the general control input for MIMO have to be accounted for. This is achieved by the proposed adaptive
systems are derived, and then it is adapted to fit the Boost converter indirect fuzzy-based scheme, tuned using a novel membership func-

1157
M. Nabipour et al. Renewable and Sustainable Energy Reviews 74 (2017) 1147–1169

Fig. 3. Conventional and Proposed Fuzzy Parameter Tuning Routines (a) Conventional Tuning Routine of Fuzzy Parameters (b) Proposed Tuning Routine of Fuzzy Parameters.

tions tuning routine. In the proposed algorithm, all the fuzzy member- must also be considered. It could easily be concluded that various
ship functions (i.e. the antecedent and consequent) are tuned synchro- membership functions shapes affect different plant differently; hence,
nously using a simple routine, resulting in a controller with more finding the optimum shape increases the design complexity.
degree of freedom and faster dynamics, compared to conventionally- To increase the degrees of dynamical freedom, improve the
tuned indirect fuzzy schemes and smoother steady states compared to dynamic response of the indirect fuzzy approach, minimize the pre-
conventional direct fuzzy controllers. implementation tuning efforts and find desirable membership func-
tions shapes, online synchronous adaptation of antecedent and con-
sequent membership functions is proposed in this research. In the
5. Proposed tuning routine and MPPT controller (realistic
presented approach the shape of the membership functions can be
approach)
modified freely. It will be exhibited in the simulations section that, by
using the proposed tuning algorithm, the dynamic and steady states
In the realistic approach, unlike the ideal form of the problem, the
responses of the controller improve significantly when compared to
plant dynamic model and state-based functions Ai ( xi ) and bi(xi) are
conventional adaptation routines. On the other hand, regardless of the
assumed to be immeasurable. Moreover Pi ( x , xi mi ), is also considered to
number of rules, the number of the variables require tuning before
be immeasurable. Hence, to design a suitable control law guaranteeing
implementation are limited to 3, minimizing the pre-implementatio-
the desirable control goals (given later in this section), Ai ( xi ), bi ( xi ) and
n(offline) computation cost. Moreover, instead of heavy optimization
Pi ( x , xi mi ) have to be estimated accurately. In this regard, the design
routines utilized in [40–42], simple algorithms are utilized to tune the
process benefited from the principles of fuzzy logic approach, the global
membership functions; hence, the online calculation burden is also
estimation theory and lyapunov's theory [96]. Based on these estima-
decreased. Free shape adaptation could also be pointed out as one of
tions, the desirable control law is generated.
the major superiorities of the proposed work over conventional tuning
Using the global estimation theory, which is the key aspect of
approaches. In this regard, the overall proposed membership functions
indirect fuzzy approach, the fuzzy logic controller can be used to
tuning approach is depicted in Fig. 3. Further explanation correspond-
estimate any bounded function. The main structure of the fuzzy
ing to the adaptation laws and the tuning procedures are given in the
controller consists of the general if-then rules, presented as “Rj ≡ if
∼j ∼j ∼j subsequent sections.
x1 ∈F1 and x2 ∈F2 …and xn ∈Fn then yj=θj, where j=1,2,…,M” and the
In this approach, to tune the membership functions, the overall
inference process related to them. Hence, the output of the fuzzy
antecedent membership functions consist of n sub-membership func-
controller is constructed as follows:
tions. Each sub-membership function is modified online until the
M optimum shape is reached. In this regard, the tuning algorithms are
∑ j =1 f j θ j M
Y= M
= g ∑ f jθj explained in this section. Using the control law ui, generated by the
∑ j =1 f j (17)
j =1 proposed indirect fuzzy approach, the state variables are moved towards
or stay in a zone with small boundaries. In other words, the overall
where:
control system will be proven to be stable, asymptotically (i.e. | x i | ≤ X0 ).
n
1 Moreover, the tracking error is also moved towards or stays in a zone
fj = ∏ μF∼i j , g = M with small boundaries about zero (i.e. |ei | ≤ E1). Similarly, the reference
i =1 ∑ j =1 f j (18)
output trajectory is proven to be tracked with minimum error (i.e.
The major contribution of the proposed design routine, besides the | ei | ≤ E0 ). Finally, the final RMS values are proven to tend towards a zone
actions taken to estimate bi ( xi ), rests in the form considered for the with small boundaries (i.e. RMS (ei ) ≤ R0 and RMS (eˆi ) ≤ R1). To reach the
membership functions. In the majority of the published documents mentioned goals, the controller design assumptions are expressed as:
regarding the online adaptation of the membership functions para-
meters, consequent membership functions are solely adapted [37]. In • Controller design assumptions:
these types of tuning algorithms, only one parameter degree of modifica-
tion, where only θ j is modified, exists. Moreover the antecedent (i) The state function Ai ( xi ) and the perturbations function Pi ( x , x mi )
membership functions have to be tuned prior to controller implementa- are assumed to be unknown and immeasurable, but limited in a
tion using heavy offline routines, where the calculation cost is directly manner so that:
proportional to the number of rules and input variables. A good example sup|Ai ( xi )| ≤ A0
is tuning an indirect fuzzy controller for a plant with 6 inputs, and 3 t >0 (19)
antecedent membership functions dedicated to each input. To design and
such a system, 54 variables have to be tuned prior to the controller
design. The parameters to be tuned rise to 63 for a system with 7 inputs. sup|Pi ( x , xi mi )| ≤ P0
t >0 (20)
Moreover, the effect of antecedent membership function shapes (i.e.
trapezoidal, triangular, Gaussian etc…) on the overall system behavior

1158
M. Nabipour et al. Renewable and Sustainable Energy Reviews 74 (2017) 1147–1169

(ii) Similarly, the control gain bi ( xi ) is assumed to be unknown and (τi )

γi = −
immeasurable in terms of magnitude and sign. This assumption is ϑi (34)
in direct contrast with the simplifying assumptions made regard-
ing the variation nature [37,39] or even the sign of the control Δi = −5.5(τi ) + 4.5 sgn(τi ), Δ′i = −5.5(τi ) + 4.5 sgn[(τi ) θi j ] (35)
gain [38], made in great share of researches discussing indirect
Δi
fuzzy approaches. αi̇ = −
2βi (36)
(iii) The reference output vector components, yd i =
[ ydi , ydi̇ , …, yid (mi −1) ]T ∈ R mi , are assumed to be bounded as j
j Δ α fˆ
|| yd i || ≤ Y0 and ||ydi(mi ) || ≤ Y1. θˆi̇ = −( i )( i )[ j i ]
(iv) The fuzzy logic output “Y” can be used to estimate any unknown Δ′i βi fˆ wi + 1 (37)
i
function denoted as “H” on a bounded set in a manner so that:
j
Δi fˆi − 2βi 2θˆi ̇
j
||H − Y || ≤ ξ [96]. In this regard, the actual state function is j
defined as: fˆi ̇ =
2αi βi (38)
Ai (xi ) = Wi *T (xi ) θ*i (xi ) + γi* (xi )
Proof of Theorem:
M
= ∑ j =1 g*f *i j (xi ) × θ*i j (xi ) + γi* (xi ) (21) To prove Theory-1 and achieve the controller design goals, the
following lyapunov's candidate function is utilized:
1 n
g* = M
∑ j =1 f *i j (22) V= ∑ Vi
i =1 (39)
where:
where:
W *iT = [W *i1 ( x i ), ... W *iM ( x i )] = g*[ f *i1 ... f *iM ] (23) M
1 1 1 ∼2 1 ∼ 2
Vi = eiT Si ei + ∑ (αi j∼fi j ∼j
+ βi j θi )2 + ϑi γ∼i 2 + εi Pi + ηi bi
θiT = [θi1 ( x i ), ... θiM ( x i )] (24) 2 j =1 2 2 2
1 1 ∼2 1 ∼ 2
Using assumption iv, estimated values of the state function are = eiT Si ei + + ϑi γ∼i 2 + εi Pi + ηi bi
given as: 2 2 2
M
1
⌢ T
Ai ( x i , θˆi ) = Wˆi ( x i , θˆi ) θˆi ( x i ) + γˆi ( x i ) (25)
+ ∑ [(αi j∼fi j )2 + (βi jθ∼i j )2 + 2αi j∼fi j βi jθ∼i j ] ≥ 0
2 j =1 (40)
M
j j while αi j and βi j are dynamic gains, defined to guarantee the overall
ˆ ˆi ( x i , θˆi ) × θˆi ( x i , θˆi ) + γˆi ( x i )
= ∑ gf
j =1
system stability and Si is the solution to the following equation:
Si Fi + Fi T Si = −Qi (41)
1
gˆ = M j
∑ j =1 fˆi (26) where Qi is a symmetrical positive definite matrix. Requirements
regarding the solutions to (41) are given in [97]. By neglecting
As a result, the estimated functions error is obtained as: subscripts and superscripts to prevent overcrowding the formulations,
∼ ⌢ the derivation of Vi is obtained as:
Ai (xi , θˆi ) = Ai (xi ) − Ai (xi , θˆi ) (27)
∼ ∼
where Vi̇ = ei̇ T Si ei + eiT Si ei̇ − ϑi γ∼i γˆi ̇ − εi Pi Pˆi̇ − ηi bi bˆi̇
⎧ M
M M ⎪ ∼2 ∼ ̇ ∼2 − 2β 2θ ̇ × θ∼+2αβ̇ × ∼ ∼
∼
Ai (xi , θˆi ) = ∑ g*f *i j (xi ) × θ*i j (xi ) − ∑ gfˆ ˆi j (xi , θˆi ) × θˆi j (xi , θˆi ) + ⎨ ∑ [2ααf ̇ − 2α 2f ̇ × f + 2ββθ f ×θ
⎪

j =1 j =1
⎩ j =1
⎫
+ γ∼i (xi , θˆi ) (28) ∼ ∼ ∼ ∼ ⎪
+ 2αβ ̇ × f × θ − 2αβ × f ̇ × θ − 2αβ × f × θ ]̇ ⎬
⎪

Theory-1: ⎭ (42)
Regarding assumptions (i-iv) corresponding to designing a suitable Combining (12) and (28), it is concluded that:
control law for MIMO plants given in (6), by using the following control
M
input, tuned using the proposed membership functions tuning routine 1 j j
ei̇ = Fi ei + Gi {vi − Ai − Pi − bi ( ){vi − [ ∑ gf ˆ ˆi (xi , θˆi ) × θˆi (xi , θˆi )]
and rules given in (31−38), the controller design goals are achieved: bˆi j =1

1 + γˆi (xi ) + Pˆi} (43)

ui = [Uˆi + UK , i ]
bˆi ( xi ) (29)
thus:
where:
⎛1⎞
Vi̇ = −eiT Qi ei + 2eiT Si Gi [vi − Ai − Pi − bi ⎜ ⎟ {vi
Uˆi = −gi ( xi ) ∑ fˆi j ( xi ) × θˆi j ( xi ) + εf ( x i )] − Pˆi ( x , x mi ) (30) ⎝ bˆi ⎠
M
UK , i = Ki (τi ) + Γi sgn(τi ) (31) − ∑ gfˆ ˆi j (xi , θˆi ) × θˆi j (xi , θˆi ) + γˆi (xi ) + Pˆi]
∼ j =1
Considering the error variables as τi = 2 ei Si Gi, γ∼i = γi − γˆ,i Pi =
∼
Pi − Pˆi, bi = bi − bˆi , the tuning rules are: M
∼2 ∼
+ ∑ ̇ ∼2 − β 2θ ̇ × θ∼ + αβ̇ × ∼
̇ × f − α 2f ̇ × f + ββθ
[αα
∼
f ×θ
M
τi
bˆi̇ ( xi ) = −
j =1]
[vi − ∑ gfˆ ˆi j (xi , θˆi ) × θˆi j (xi , θˆi ) + γˆi (xi ) + Pˆi] ∼∼ ∼ ∼ ∼ ̇
ηi [bˆi ( xi )] j =1 (32) j × θ ]̇ − ϑiγiγi
+ αβḟ θ − αβf ̇ × θ − αβf ∼ ˆ̇ − εi Pi Pˆi ̇ − η bi bˆi (44)
i

M j j
(τ ) By defining a new variable as L i = vi − ˆ ˆi
∑ j =1 gf (xi , θˆi ) × θˆi (xi , θˆi )
Pˆi ̇ ( xˆ, xˆ mi ) = − i
εi (33) +γˆ (xi ) + Pˆi (44) can be modified to:
i

1159
M. Nabipour et al. Renewable and Sustainable Energy Reviews 74 (2017) 1147–1169

1 relations could also be concluded:

ei̇ = Fi ei + Gi {vi − Ai − Pi − bi ( )(L i )}
bˆi (45)
Vi̇ ≤ −Kv Vi + Kv V ′d (55)
And hence:
Vi (t ) ≤ max(V (0), V ′d ) (56)
1
Vi̇ = −eiT Qi ei + 2eiT Si Gi [vi − Ai − Pi − bi ( ) L i ] where
bˆi
M λ min (Qi )
+ ∑ [αα̇ × ∼f 2 ∼ ̇ ∼2 − β 2θ ̇ × θ∼ + αβ̇ × ∼
− α 2f ̇ × f + ββθ
∼
f ×θ
Kv =
λ (57)
j =1
∼∼ ∼ ∼ λ = max [λ max (Si )] (58)
+ αβḟ θ − αβf ̇ × θ − αβf × θ ]̇
∼ ∼ 1
∼ ∼
−ϑi γ∼i γˆi ̇ − εi Pi Pˆi̇ − ηi bi bˆi̇ V ′d = Kv ∑ max(αf + βθ ) + Kv { max [γ∼i 2] + Vd
(46) 2ηi (59)
which can be modified to:
max(V (0), V ′d )
∼ | e| ≤
b λ min (Pi ) (60)
Vi̇ = −eiT Qi ei + τi [vi − Ai − Pi − ( i ) L i − L i ]
bˆi
M max(V (0), V ′d )
| x i | ≤ | yd | + | e| ≤ Y0 +
+ ∑ [αα̇ × ∼f 2 ∼ ̇ ∼2 − β 2θ ̇ × θ∼ + αβ̇ × ∼
− α 2f ̇ × f + ββθ
∼
f ×θ λ min (Pi ) (61)
j =1
∼∼ ∼ ∼ lim(λ min (S )|e| ≤ eT Se) ≤ V ′d
+ αβḟ θ − αβf ̇ × θ − αβf × θ ]̇ t →∞ (62)
∼ ∼
−ϑi∼γi∼γi − εi Pi Pˆi̇ − ηi bibˆi̇ (47) lim |e| ≤
V ′d
t →∞ λ min (Si ) (63)
M
∼2 ∼ ̇ ∼2 − β 2θ ̇ × θ∼ + αβ̇ × ∼ ∼
=−eiT Qi ei + ∑ ̇ × f − α 2f ̇ × f + ββθ
[αα f ×θ t Vd
j =1
lim
t →∞
∫0 (| e|2 ) dt ≤
λ min (Qi ) (64)
∼∼ ∼ ∼
+ αβḟ θ − αβf ̇ × θ − αβf × θ ]̇
⎡ 1 t ⎤ Vd
∼ ∼ ∼ L ⇒RMS (e) = lim ⎢ ∫0 (| e∼|2 ) dt ⎥ ≤
−τi Ai + γ∼i (−ϑi γˆi ̇ − τi ) + Pi (−εi Pˆi̇ − τi ) + bi (−ηi bˆi̇ − τi i ) t →∞ ⎣ t ⎦ λ min (Qi ) (65)
bˆi (48)
As a result, the design goals are clearly proven. To summarize, the
Using Eqs. (29), (31)–(38), the following equation s concluded from variables that could be modified to tighten the bounds surrounding
(48): various parameters are expressed in Table 10.
∼2 ∼2 ∼∼
Vi̇ = −eiT Qi ei + ∑ [θ (2ββ̇ ) + f (2αα ̇ ) + f θ (2αβ̇ + 2αβ ̇ − g* (τ ))
∼ 6. Simulations and modeling
+ θ (−g* (τ ) fˆ − 2αβfˆ ̇ − 2β 2θˆ )̇
∼
+ f (−2α 2fˆ ̇ − 2αβθˆ ̇ − g* (τ )( ˆf w + 1) θˆ )] To examine the applicability of the proposed scheme in closed loop
control structures, especially MPPT tasks, some simulations were
(49)
performed on the PV-module presented in Fig. (1b). In these simula-
The membership functions are defined in a way that for all i > 0 we tions, the PV-module is an NE165U1 module introduced in [98,99].
could have 1≤ gi,gi*≤10. Hence Eq. (49) could be approximated as: The corresponding nonlinear characteristic curves are depicted in Figs.
∼2 ∼2 ∼∼ (4.a and .b). In this regard, the utilized Boost converter parameters are
Vi̇ ≤ − eiT Qi ei − Ki |τi ] + ∑ [θ (2ββ̇ ) + f (2αα ̇ ) + f θ (Δ + 2αβ̇ + 2αβ )̇
as [RL, L, C, RC] =[0.01Ω,0.01 H, 10μF, 25Ω], In these simulations, the
∼ ∼
+ θ (Δfˆ − 2αβfˆ ̇ − 2β 2θˆ )̇ + f (−2α 2fˆ ̇ − 2αβθˆ ̇ + Δ′( ˆf w + 1))] (50) following state equations are utilized:
where 1
Iṡ = (−RL + Vs − Vo (1 − D )) + dB1
L (66)
⎧ − (τ ) (τ ) ≥ 0 ⎧ − (τ ) (τ ) θˆ ≥ 0
Δ=⎨ , Δ′ = ⎨ 1
⎩− 10(τ ) (τ ) ≤ 0 ⎩− 10(τ ) (τ ) θˆ ≤ 0 (51) VȮ = ((1 − D ) Is − Io ) + dB2
C (67)
Equivalently:
Assuming state vectors as x = [x ] = [Is], e = e = [ yd − y] =
Δ = −5.5(τ ) + 4.5 sgn[(τ )](τ ) [IMPP* − Is] , regarding Eqs. (38) and (39), in order for the input current
Δ′ = −5.5(τ ) + 4.5 sgn[(τ ) θˆ](τ ) (52) to follow the reference MPP current accurately, with the IMPP*
generator and the MPPT controller defined as depicted in Fig. (4c),
j
By substituting Eqs. (35)–(38) in (50), if βi is selected to be a the following control law was applied to the converter as the duty cycle:
∼2 ∼2
constant, it is logical to assume f < < 1 and hence F (2αα̇ ) could be
1
neglected. As a result, we have: u=D= ˆ
(v + Γsgn (e) − w (Is ) ∑ Fˆ j (Is) × θˆ j (Is) − γˆ (Is) − Pˆ (Is, Iṡ )
bi
Vi̇ ≤ Γ ′i + ∑ [F∼2 (2αα̇ )] = − eiT Qi ei + Vd (53) (68)
∼2
where Vd = −Ki |τi ] + ∑ [F (2αα ̇ )] ≅ −Ki |τi ] < 0 . Table 10
Considering − eiT Qi ei ≤ −λ min (Qi )|| e||2 ), with the minimum Eigen Influence of Variables on System Bounds.
value denoted as λ min (.), the following is easily concluded:
| ei || | xi | lim (|e|) RMS (e)
Vi̇ ≤ −λ min (Qj )(|| e||2 ) + Vd ≤ 0 (54) t→∞

Parameters to max(V (0), V ′d ) Y0, max(V (0), V ′d ) V′d V′d

Considering (40) and (54) where it is found that
decrease
{ ∀ i|vi ≥ 0andvi̇ < 0}, as described in [95], the overall system is proven Parameters to increase λ min (Si ) λ min (Si ) λ min (Si ) λ min (Qi )
to be asymptotically stable and lim ei (t ) = 0 . Moreover, the following
t →∞

1160
M. Nabipour et al. Renewable and Sustainable Energy Reviews 74 (2017) 1147–1169

Fig. 4. : The Control Setup and PV Characteristic Curves (a) P-I characteristic curve of the utilized PV-module (b) I-V characteristic curve of the utilized PV-module (c) Control Setup (d)
θij and fij.

Table 11 of the membership function, defined using structural parameters which

Adaptive fuzzy rules of the proposed controller. are usually 3 for each membership function. Hence, by altering these
structural parameters the membership degree of a specific state
xi xi ∈ Fi1 xi ∈ Fi 2 xi ∈ Fi 3
variable, at a certain magnitude could be modified. Moreover, since
Yi Yi = θi1 Yi = θi 2 Yi = θi 3
these membership degrees directly affect the output of fuzzy rules, their
determination prior to controller implementation is of vital impor-
̇ *, Γˆ = 1000 and other parameters are deter-
where v = 50000e + IMPP tance.
mined based on (31)−(38). In the simulations performed, the fuzzy In conventional controller design procedures determining suitable
membership functions and rules presented in Fig. (4d) and Table 11 membership functions, as mentioned in numerous occasions through-
are employed. Moreover, without loss of generality, one could design out this research is a time-consuming task with high calculation
the controller considering ηi = ϑi = εi = 0.01. burden. On the other hand, in the proposed scheme, due to continuous
The utilized membership functions for the proposed fuzzy-based adaptation of membership functions, when the corresponding rules are
scheme are presented in Fig. (4c). The height of the antecedent and triggered, the membership degrees constantly adapt and tend towards
horizontal position of the consequent membership functions vary optimum values. Hence, the initial structural parameter values will not
based on Eqs. (34) and (36)–(38) to guarantee system stability or have a significant effect on the performance of the overall control
minimum tracking error. scheme. These functions along with the output centroids are adapted
In design and implementation of the proposed scheme some points using Eqs. (32)–(38) based on the instantaneous values of the state
must be considered. In this regard, when designing a conventional variables.
j j
indirect fuzzy-based scheme, the antecedent membership functions are Based on Eqs. (37) and (38), it is easily concluded that θ̂i and fˆ i
static. Due to these static membership functions, the membership remain constant when the fuzzy rule corresponding to fi j is not
j j j
degree of a specific state variable, at a certain magnitude, will always triggered. Meaning that θ̂i and fˆi will adapt if and only if fˆi possesses
remain stationary. This membership degree is determined by the shape a non-zero magnitude. Hence, based on the proposed membership

1161
M. Nabipour et al. Renewable and Sustainable Energy Reviews 74 (2017) 1147–1169

function structure, depicted in Figs. (3b, 4d), it can easily be concluded 3 at all conditions. This considerable reduction in pre-implementation
that the maximum number of antecedent membership functions which calculation effort is another advantage of the proposed scheme over
are simultaneously modified is limited to 2 for each state variable. The conventional methods.
rest of the antecedent membership functions are excluded from the In this section, five sets of simulations were performed. Initially, the
adaptation procedure. This fact significantly reduces the control behavior of the PV-module controlled using the proposed scheme are
system's computation load compared to the schemes presented in examined and the proposed MPPT algorithm is compared with
[37], where all the membership functions must be monitored con- conventional fuzzy-based MPPT scheme. In the second set, sinusoidal
stantly. variations, in the third set “one day temperature and irradiance level”
Besides reduction in the online calculation cost, due to negligible variations are compared using linear irradiance variation tests. In the
effect of the initial settings of the proposed approach on the controller 4th set of simulations, the proposed scheme is compared with the
performance, the pre-implementation calculation effort can also be conventional indirect fuzzy controller in boost current tracking objec-
reduced using the proposed scheme. This is due to the dependence of tives. The last set of simulations tries to evaluate the generalization
the characteristics of the proposed scheme to only 3 parameters, capability of the proposed scheme.in this set of tests the performance of
namely ηi , αi and βi , while in the conventional indirect fuzzy approaches the proposed scheme in tracking the MPP implemented on several
[37], the controller behavior is affected by 3 parameters for each types of multi-component or in other words Multi-Junction (MJ) PV
membership function, namely the membership functions structural models is examined.
parameters. These parameters increase exponentially as the number of T.1 Irradiance variations:
required rules increase; while for the proposed scheme, the number of In this test, the purpose was to apply various possible linear
variables with pre-implementation tuning requirements are constantly irradiance trajectory variations to the PV-module. Hence, the irradi-

Fig. 5. : Irradiation Variation Test (a): The Applied Irradiance Trajectory (b): Output Current of the PV-Module (c): Irradiance Variation Test Results Zone-A (d): Irradiance Variation
Test Results Zone-B.

1162
M. Nabipour et al. Renewable and Sustainable Energy Reviews 74 (2017) 1147–1169

ance trajectory, depicted in Fig. (5.a) was applied to the PV-setup. the initial drawn current from the PV-arrays was Is=2 A. With the first
Moreover, in Fig (5.b- d), the MPPT performance of the proposed drop, the current was decreased to about 1 A, and in the ramp starting
scheme and the conventional direct fuzzy-based MPPT routines are at t=100 ms, reached 5 A. Throughout the irradiance variations, as
demonstrated. demonstrated in Fig. (5b), the current was varied in the interval of [1 A,
As depicted in Fig. (5a), at t=25 ms, after the OP of the PV-module 5 A]. In Figs. (5c and d) the characteristic curves corresponding to
has reached steady states, with a steep drop, the irradiance trajectory zones A and B corresponding to both approaches are plotted respec-
was set to S=250 W/m2 form S=500 W/m2. At t=40 ms, another step tively. Based on these figures, the oscillatory behavior of the conven-
in the irradiance increased this level to S=750 W/m2. Two more steep tional fuzzy-based MPPT schemes in tracking the IMPP* is clearly
variations dropped and rose the irradiance levels at t=60 ms and observable. This fact is also present in the “Active Power” plots
t=80 ms, to S=250 W/m2 and S=500 W/m2 respectively. From this depicted in Fig. (5c, d). On the other hand, the proposed scheme
point forth, ramp variations were applied to the PV-module, raising the tracked the reference irradiance level with minimum error. Therefore,
level to a maximum of S=1000 W/m2 and returning it back to the drawn current followed the reference smoothly and the correspond-
S=500 W/m2. The initial ramp starts at t=100 ms and the second ing power followed the MPP at all instances.
ramp initiated at t=200 ms. Using this type of variation, apart from T.2 Sinusoidal irradiance variation test:
dynamic behavior examinations, the steady states behaviors of the In the next test, the irradiance trajectory is varied sinusoidally. The
mentioned schemes are also examined. applied sinusoidal trajectory is demonstrated in Fig. (6a). The varia-
Based on Fig. (5b), the initial Boost input current or in other words, tions start at t=0.07 s with a magnitude equal to 250 W/m2 increasing

Fig. 6. : Sinusoidal Variations test results (a): Applied Irradiance trajectory (b): Sinusoidal Test current tracking (c): Sinusoidal Test results of Zone A (d): Sinusoidal Test results of zone
B.

1163
M. Nabipour et al. Renewable and Sustainable Energy Reviews 74 (2017) 1147–1169

the irradiance level to 1000 W/m2. As a result of this signal, the should be noted that sudden drops and rises in the irradiance level,
irradiance level varies between 500 W/m2 and 1000 W/m2 sinosoid- generally used to model clouds blocking sunlight have also been
ally. The test results are demonstrated in Fig. 6. considered. The applied irradiance and temperature are depicted in
Similar to the ROPP test, large amounts of current ripples is Fig. (7a).
exhibited by the conventional fuzzy controller's results. Yet, the super- As depicted in Fig. (7a) temperature variations follow similar
iority of the proposed controller is again proven since the current principles. The temperature trajectory initially starts from minimum
trajectory was tracked with the least amount of oscillations possible by values, generally modeling early morning conditions. These variations
the proposed controller and as a result, negligible active power start from the minimum temperature level of 300 K at t=6 H and
oscillations are observed. increase until it reaches T=350 K at t=13 H. Returned to T=310 K at
T.3 One day irradiance and temperature trajectory: t=18.5 H, the temperature trajectory was designed in a manner so that
In the third set of simulations, besides variations in irradiance it exhibits sinusoidal characteristics throughout the simulation time.
levels, the behavior of controllers considering general variations in the The behavior of the output current of the PV module in both cases of
temperature levels throughout one single day are also examined. The utilizing conventional FL or the proposed approach is depicted in Fig.
utilized data has been collected at a site near Madrid in Spain at March (7b).
[98]. In this case, by considering the simulation length equivalent to In Figs. (7c, d) the zoomed plots of the operational behavior of the
the time interval of the day where sunlight is present, the irradiance examined controllers respectively corresponding to Zones A and B are
level is initially set to its minimum value. As the time passes, this level demonstrated. Similar to the ROPP and sinusoidal tests, high magni-
is increased until it reaches its maximum of S=950 W/m2. Then, it tudes of oscillations are demonstrated using the conventional direct
reduces until the minimum irradiance is reached again at t=18 H. It fuzzy approach. Using the proposed scheme the oscillations in the

Fig. 7. : One Day Irradiance and Temperature Trajectory (a): Applied Temperature and Irradiance Trajectory (b): Output Current r of the PV-Module (c): Output Current and Active
Power of the PV-Module Zone A (d): Output Current and Active Power of the PV-Module Zone B.

1164
M. Nabipour et al. Renewable and Sustainable Energy Reviews 74 (2017) 1147–1169

drawn current are minimized. Diminishing the ripple magnitudes in Table 12

the drawn current leads to smoother active power from the PV module. Control Comparison Results.
These can be confirmed using Figs. (7c, d).
Instance of Reference
T.4 Tracking reference current performances: Variation
Due to various reasons accurate comparison of the proposed and Tuning approach t=0 ms t=30 ms t=40 ms t=50 ms t=60 ms
conventional membership function tuning approaches cannot be
Conventional Rising time 8 ms 2 ms 3 ms 3 ms 3 ms
accomplished in the performed MPPT tests. Compensation of the
Approach Settling time 22 ms 3 ms 3 ms 11 ms 3 ms
effects of inaccurate function estimations using large gains, actuator [37]
saturations caused by large control signals and limited over/under-
shoot due to nonlinear PV-array characteristics could be counted as the Proposed Rising time 8 ms 1 ms 2 ms 2 ms 2 ms
main reasons. Therefore, reference tracking performance by the Boost- Approach Settling time 8 ms 1 ms 2 ms 2 ms 2 ms.

converter, fed with a fixed input voltage might be a suitable way to

directly compare estimation capabilities of the proposed scheme and
of these modules is presented in [103]. In this regard, I-V and P-V plots
the conventional approaches such as [37] in tuning continuous
corresponding to a 10 cm×10 cm single PV cell, where S=1000 W/m2
functions. In this regard, the reference current signal, as depicted in
and T=300 K are respectively presented in Figs. (9a and b).
Fig. 8 was fed to the Boost controller. The control parameters
As exhibited in Figs. (9a and b) and ref. [103], in these MJ-PV
corresponding to this simulation are set to VDC=300 V,
modules, despite acceptable current density, a rather low maximum
v = 5000e + IMPP* and Γ̂ = 500 .
power is presented. For nc-Si, CdTe and a-Si/nc-Si the expressed
The results of this test are depicted in Fig. 8. From Fig. 8 it is
maximum powers are 2.2 W, 1.4 W and 1.2 W respectively. To improve
evident that the control performance was followed by overshoot/
the maximum transmittable power so that the modules could be
undershoots when using the conventional indirect fuzzy-based ap-
practically implementable in realistic load conditions, for each MJ-
proach; while utilizing the proposed scheme, the tracking performance
PV type, the PV module is considered to be constructed using a
was faster, smoother and with minimum error.
combination of 80 cells connected in series. In this regard, the modified
On the other hand, while the conventional indirect fuzzy approach
I-V and P-V plots corresponding to the modified multi-cell MJPV
[37] exhibited notable overshoot/undershoots of about (150%, 5% and
modules are respectively presented in Figs. (10a and b).
10%) respectively at t=[15 ms, 35 ms and 55 ms], no overshoots were
Based on Figs. (10a and b), significant improvements in the
observed in the performance of the proposed scheme. Moreover,
maximum transmittable power is observed. The performance of the
comparison between rise/settling time of the two controllers in various
proposed controller in comparison to the conventional FL was again
reference variations are given in Table 12.
evaluated in tracking the MPP of the resulting multi-cell MJ-PV
As mentioned before, by synchronously tuning the antecedent and
modules subjected to irradiance level variations similar to the varia-
consequent membership functions, by exhibiting faster and more
tions applied in section (V) Fig. (5a). It should be noted that the
accurate tracking, the dynamic performance of the controller has
variations in the overall performance of both examined schemes
improved. The superior performance of the proposed scheme in
corresponding to changes in the constructing MJ-PV materials is not
dynamic states by reducing the rise/settling time and in steady states
considerable and only the magnitudes of steady states and transient
by minimizing the tracking error are clearly evident using Table 12 and
oscillations of the conventional FL seem to experience some changes.
Fig. 8. As a result of elimination of the overshoots/undershoots, the
In this regard, the diagram of the MJ-PV module output “Active Power”
values for the rise/settling time in the proposed scheme have become
is depicted in Fig. 11. In this figure the steady states and dynamic zones
equal and have been minimized compared to the scheme presented in
are respectively distinguished as “Zone-A” and “Zone-B”.
[37].
Initially the AIST 11.4% nc-Si [100] is subjected to irradiance
T.5 Generalization of the proposed scheme:
variations. Zoomed plots of the active powers are presented in Fig. (12a
To confirm generalization of the proposed scheme, simulations on
and b). Similar to the results given in section (V.T1), for the Zone-A
other types of PV modules, with the proposed and conventional FL
results, the proposed scheme followed the MPP with the least error
schemes acting as the MPPT controllers were performed. In these
possible while the MJ-PV module controlled by the conventional FL
simulations, the efforts were put into testing the proposed scheme on
experienced notable power oscillations as expected. Similar results are
different PV modules constructed using different materials. These
concluded from Fig. (12b) where the zoomed plot of Zone-B is
modules are referred to as multi-junction PV (MJ-PV) modules. In
depicted. As a result, the absolute superiority of the proposed scheme
this regard, AIST Nano-crystalline silicon (nc-Si) [100], First Solar,
in tracking the MPP over the conventional FL is clearly observable from
monolithic Cadmium telluride (CdTe) [101] and TEL Solar, Trubbach
Fig. 12.
Labs amorphous silicon/hydrogen alloy with Nano-crystalline silicon
In Figs. (13 and 14) the results of MPPT control applied to First Solar
alloy (a-Si/nc-Si) [102] PV modules were subjected to test conditions.
21% CdTe and Tel Solar 12.2% a-Si/nc-Si are respectively plotted.
The corresponding current density as a function of the output voltage

Fig. 8. : Reference Tracking Results (a): Current Tracking Performance Test Results (b): Reference Tracking Errors.

1165
M. Nabipour et al. Renewable and Sustainable Energy Reviews 74 (2017) 1147–1169

Fig. 9. Output characteristics of various MJ-PV cells (a) I-V plot of various MJ-PV cells (b) P-V plot of various MJ-PV cells.

Fig. 10. Output characteristics of various multi-cell MJ-PV cells (a) I-V plot of various multi-cell MJ-PV modules (a) P-V plot of various multi-cell MJ-PV modules.

various weather conditions (irradiance and temperature variations).

7. Conclusions

A review over conventional MPPT routines is presented in this

paper. In this review, after defining minimum MPPT expectations,
various MPPT methods were compared in terms of the ability to cover
all the defined objectives at the same time. At the end of the review it
was concluded that among the available MPPT control methods, only
an “antecedent-consequent adaptive” fuzzy based scheme has the
Fig. 11. The MJ-PV module output “Active Power”. potential to fulfill all the design objectives at the same time. Hence, a
novel membership functions tuning routine to accomplish optimum
The obtained results are similar to the previous results confirming MPPT objectives are introduced in this research. On the contrary to
the absolute superiority of the proposed scheme over conventional FL available tuning routines, the antecedent and consequent membership
scheme. In this regard, with the MPP being tracked with the minimum functions of the fuzzy-based system are tuned synchronously using a
error possible, it could be concluded that variations in the structure, computationally light algorithm. To examine the applicability of the
materials and formation of the PV module has minimum effect on the proposed scheme, simulations were performed. In these simulations,
performance of the proposed scheme. Based on these conclusions, the the general MPPT field tests, such as the ROPP test, sinusoidal and
proposed scheme can operate desirably in a wide range of PV module “one day temperature and irradiance level variation” and the current
types constituted using various types of materials and subjected to tracking tests were performed in MATLAB-SIMULINK environment.

Fig. 12. Active Power plot corresponding to AIST 11.4% nc-Si (a) Active Power plot corresponding to Zone-A of AIST 11.4% nc-Si multi-cell MJ-PV module (b) Active Power plot
corresponding to Zone-B of AIST 11.4% nc-Si multi-cell MJ-PV module.

1166
M. Nabipour et al. Renewable and Sustainable Energy Reviews 74 (2017) 1147–1169

Fig. 13. Active Power plot corresponding to First Solar 21% CdTe (a) Active Power plot corresponding to Zone-A of First Solar 21% CdTe multi-cell MJ-PV module (b) Active Power plot
corresponding to Zone-A of First Solar 21% CdTe multi-cell MJ-PV module.

Fig. 14. Active Power plot corresponding to Tel Solar 12.2% a-Si/nc-Si (a) Active Power plot corresponding to Zone-A of Tel Solar 12.2% a-Si/nc-Si multi-cell MJ-PV module (b) Active
Power plot corresponding to Zone-A of Tel Solar 12.2% a-Si/nc-Si multi-cell MJ-PV module.

In all the employed tests, the proposed MPPT routine was compared [7] Reisi AR, Moradi MH, Jamasb S. Classification and comparison of maximum
power point tracking techniques for photovoltaic system: a review. Renew Sustain
with conventional direct fuzzy MPPT routine. The conventionally tuned Energy Rev 2013;19:433–43.
adaptive indirect fuzzy-based MPPT routines were also compared with [8] Mellit A, Kalogirou SA. MPPT-based artificial intelligence techniques for photo-
the proposed approach in controlling a reference trajectory. As voltaic systems and its implementation into field programmable gate array chips:
review of current status and future perspectives. Energy 2014;70:1–21.
demonstrated in the simulation results, tuning the antecedent and [9] Esram T, Chapman PL. Comparison of photovoltaic array maximum power point
consequent membership functions improves the tracking performance tracking techniques. IEEE Trans Energy Convers EC 2007;22(2):439.
of the controller, in terms of steady states and dynamic characteristics. [10] Tyugu E. Algorithms and architectures of artificial intelligence. Netherlands: IOS
Press; 2007.
As a result, the ripples exhibited by the conventional fuzzy approach in
[11] Kalogirou SA. Artificial intelligence in energy and renewable energy systems. New
tracking the MPP trajectory were minimized. In this regard, the active York: Nova Science Publisher; 2007.
power oscillations exhibited by the fuzzy-based schemes were also [12] Cirrincione M, Pucci M, Vitale G. Neural MPPT of variable-pitch wind generators
with induction machines in a wide wind speed range. IEEE Trans Ind Appl
minimized. Moreover, the dynamic response was also improved
2013;49(2):942–53.
compared to the conventionally tuned adaptive fuzzy approach. The [13] Prakash J, Sahoo SK, Karthikeyan SP, Raglend IJ. Design of PSO-Fuzzy MPPT
results confirm the superiority of the proposed tuning routine in terms controller for photovoltaic applicationPower Electronics and Renewable Energy
of the active power and current oscillations, rising time, settling time Systems. India: Springer; 2015. p. 1339–48.
[14] Daraban S, Petreus D, Morel C. A novel MPPT (maximum power point tracking)
and over/undershoots, compared with conventional direct and indirect algorithm based on a modified genetic algorithm specialized on tracking the global
fuzzy MPPT routines. maximum power point in photovoltaic systems affected by partial shading. Energy
2014;74:374–88.
[15] Salah CB, Ouali M. Comparison of fuzzy logic and neural network in maximum
References power point tracker for PV systems. Electr Power Syst Res 2011;81(1):43–50.
[16] Chekired F, Mellit A, Kalogirou SA, Larbes C. Intelligent maximum power point
[1] Wei Haokun, Liu Jian, Yang Biao. Cost-benefit comparison between Domestic trackers for photovoltaic applications using FPGA chip: a comparative study. Sol
Solar Water Heater (DSHW) and Building Integrated Photovoltaic (BIPV) systems Energy 2014;101:83–99.
for households in urban China. Appl Energy 2014;126:47–55. [17] Laudani A, Fulginei FR, Salvini A, Lozito GM, and Mancilla-David F.
[2] Zhang W, Zhu R, Liu B, Ramakrishna S. High-performance hybrid solar cells Implementation of a neural MPPT algorithm on a low-cost 8-bit microcontroller.
employing metal-free organic dye modified TiO2 as photoelectrode. Appl Energy In: Proceedings of international symposium on IEEE power electronics, electrical
2012;90:305–8. drives, automation and motion (SPEEDAM); 2014 p. 977–81.
[3] Liu Yali, Li Ming, Ji Xu, Luo Xi, Wang Meidi, Zhang Ying. A comparative study of [18] Kofinas P, Dounis AI, Papadakis G, Assimakopoulos MN. An Intelligent MPPT
the maximum power point tracking methods for PV system. Energy Convers controller based on direct neural control for partially shaded PV system. Energy
Manag 2014;85:809–16. Build 2015;90:51–64.
[4] Punitha K, Devaraj D, Sakthivel S. Artificial neural network based modified [19] Messalti S, Harrag A, Loukriz A. A new variable step size neural networks MPPT
incremental conductance algorithm for maximum power point tracking in controller: review, simulation and hardware implementation. Renew Sustain
photovoltaic system under partial shading conditions. Energy 2013;62:330–40. Energy Rev 2017;68:221–33.
[5] Xiao Xi, Huang Xuanrui, Kang Qing. A hill-climbing-method-based maximum- [20] Hafiz F, Abdennour A. An adaptive neuro-fuzzy inertia controller for variable-
power-point-tracking strategy for direct-drive wave energy Converters. Ind speed wind turbines. Renew Energy 2016;92:136–46.
Electron, IEEE Trans 2016;63(1):257–67. [21] Premkumar K, Manikandan BV. Adaptive Neuro-Fuzzy Inference System based
[6] Al-Amoudi A, Zhang L. Optimal control of a grid-connected PV system for speed controller for brushless DC motor. Neurocomputing 2014;138:260–70.
maximum power point tracking and unity power factor. In: Proceedings of the [22] Chaouachi A, Kamel RM, Nagasaka K. A novel multi-model neuro-fuzzy-based
seventh international conference on power electronics and variable speed drives MPPT for three-phase grid-connected photovoltaic system. Sol Energy
(Conference Publ No. 456);1998 p. 80–5. 2010;84(12):2219–29.
[23] Anguera X, Shinozaki T, Wooters C, Hernando J. Model complexity selection and

1167
M. Nabipour et al. Renewable and Sustainable Energy Reviews 74 (2017) 1147–1169

cross-validation EM training for Robust speaker diarization. In: Proceedings of power generation. IEEE Trans Ind Electron 2015;62(2):1047–58.
ICASSP; 2007. [56] Yang H, Zhou W, Lu L, Fang Z. Optimal sizing method for stand-alone hybrid
[24] Gao Z, Ming F, Hongling Z. Bagging neural networks for predicting water solar–wind system with LPSP technology by using genetic algorithm. Sol Energy
consumption. J Commun Comput 2005;2(3):19–24, [Serial No. 4]. 2008;82(4):354–67.
[25] Yu L, Lai KK, Wang S. Multistage RBF neural network ensemble learning for [57] Larbes C, Cheikh SA, Obeidi T, Zerguerras A. Genetic algorithms optimized fuzzy
exchange rates forecasting. Neurocomputing 2008;71(16):3295–302. logic control for the maximum power point tracking in photovoltaic system.
[26] Li X, Rakkiyappan R, Balasubramaniam P. Existence and global stability analysis Renew Energy 2009;34(10):2093–100.
of equilibrium of fuzzy cellular neural networks with time delay in the leakage [58] Panda A, Pathak MK, Srivastava SP. Fuzzy intelligent controller for the maximum
term under impulsive perturbations. J Frankl Inst 2011;348(2):135–55. power point tracking of a photovoltaic module at varying atmospheric conditions.
[27] Kuo KY, Lin J. Fuzzy logic control for flexible link robot arm by singular J Energy Technol Policy 2011;1(2):18–27.
perturbation approach. Appl Soft Comput 2002;2(1):24–38. [59] Bendib B, Krim F, Belmili H, Almi MF, Boulouma S. Advanced fuzzy MPPT
[28] Lien CH, Yu KW, Chang HC, Chung LY, Chen JD. Robust reliable guaranteed cost controller for a stand-alone PV system. Energy Procedia 2014;50:383–92.
control for uncertain T‐S fuzzy neutral systems with interval time‐varying delay [60] Chiu CS, Ouyang YL. Robust maximum power tracking control of uncertain
and linear fractional perturbations. Optim Control Appl Methods photovoltaic systems: a unified TS fuzzy model-based approach. IEEE Trans
2015;36(1):121–37. Control Syst Technol 2011;19(6):1516–26.
[29] Wang Huanqing, Liu Xiaoping, Liu Peter Xiaoping, Li Shuai. Robust adaptive [61] El Khateb A, Rahim NA, Selvaraj J, Uddin MN. Fuzzy-logic-controller-based
fuzzy fault-tolerant control for a class of non-lower-triangular nonlinear systems SEPIC converter for maximum power point tracking. IEEE Trans Ind Appl
with actuator failures. Inform Sci 2016;336:60–74. 2014;50(4):2349–58.
[30] Guenounou Ouahib, Dahhou Boutaib, Chabour Ferhat. Adaptive fuzzy controller [62] Ocran TA, Cao J, Cao B, Sun X. Artificial neural network maximum power point
based MPPT for photovoltaic systems. Energy Convers Manag 2014;78:843–50. tracker for solar electric vehicle. Tsinghua Sci Technol 2005;10(2):204–8.
[31] Eltawil MA, Zhao Z. MPPT techniques for photovoltaic applications. Renew [63] Bahgat ABG, Helwa NH, Ahmad GE, El Shenawy ET. Maximum power point
Sustain Energy Rev 2013;25:793–813. traking controller for PV systems using neural networks. Renew Energy
[32] Suganthi L, Iniyan S, Samuel AA. Applications of fuzzy logic in renewable energy 2005;30(8):1257–68.
systems–a review. Renew Sustain Energy Rev 2015;48:585–607. [64] Ramaprabha R, Mathur BL, Sharanya, M. Solar array modeling and simulation of
[33] Farhat M, Barambones O, Sbita L. Efficiency optimization of a DSP-based MPPT using neural network. In: Proceedings of IEEE international conference on
standalone PV system using a stable single input fuzzy logic controller. Renew control, automation, communication and energy conservation, INCACEC 2009;
Sustain Energy Rev 2015;49:907–20. 2009. p. 1–5.
[34] Mendel Jerry M. Uncertain Rule-Based Fuzzy Logic System: Introduction and [65] Kassem AM. MPPT control design and performance improvements of a PV
New Directions. Upper Saddle River, New Jersey: Prentice Hall PTR; 2001. generator powered DC motor-pump system based on artificial neural networks.
[35] Boulkroune Abdesselem. A fuzzy adaptive control approach for nonlinear systems Int J Electr Power Energy Syst 2012;43(1):90–8.
with unknown control gain sign. Neurocomputing 2015. [66] Kamarzaman NA, Tan CW. A comprehensive review of maximum power point
[36] Lin C-K, Wang S-D. Robust self-tuning rotated fuzzy basis function controller for tracking algorithms for photovoltaic systems. Renew Sustain Energy Rev
robot arms. IEEE Proc Control Theory Appl 1997;144(4):293–8. 2014;37:585–98.
[37] Barkat Said, Tlemçani Abdelhalim, Nouri Hassan. Noninteracting adaptive control [67] Mellit A, Kalogirou SA. MPPT-based artificial intelligence techniques for photo-
of PMSM using interval type-2 fuzzy logic systems. Fuzzy Syst, IEEE Trans voltaic systems and its implementation into field programmable gate array chips:
2011;19(5):925–36. review of current status and future perspectives. Energy 2014;70:1–21.
[38] Li HX, Tong S. A hybrid adaptive fuzzy control for a class of nonlinear MIMO [68] Joshi P, Arora S. Maximum power point tracking methodologies for solar PV
systems. Fuzzy Syst, IEEE Trans 2003;11(1):24–34. systems–a review. Renew Sustain Energy Rev 2016.
[39] Boulkroune Abdesselem, M'saad Mohammed. On the design of observer-based [69] Kermadi M, Berkouk EM. Artificial intelligence-based maximum power point
fuzzy adaptive controller for nonlinear systems with unknown control gain sign. tracking controllers for photovoltaic systems: comparative study. Renew Sustain
Fuzzy Sets Syst 2012;201:71–85. Energy Rev 2017;69:369–86.
[40] M.J , Deng C. Online tuning of fuzzy inference systems using dynamic fuzzy Q- [70] Messai A, Mellit A, Guessoum A, Kalogirou SA. Maximum power point tracking
learning. Syst, Man, Cybern, Part B: Cybern, IEEE Trans 2004;34(3):1478–89. using a GA optimized fuzzy logic controller and its FPGA implementation. Sol
[41] Jou JM, Chen PY, Yang SF. An adaptive fuzzy logic controller: its VLSI Energy 2011;85(2):265–77.
architecture and applications. Very Large Scale Integr (VLSI) Syst, IEEE Trans [71] Ishibuchi H, Yamamoto T. Fuzzy rule selection by multi-objective genetic local
2000;8(1):52–60. search algorithms and rule evaluation measures in data mining. Fuzzy Sets Syst
[42] Oh S, Pedrycz W. Identification of fuzzy systems by means of an auto-tuning 2004;141(1):59–88.
algorithm and its application to nonlinear systems. Fuzzy Sets Syst [72] Castillo L, González A, Pérez R. Including a simplicity criterion in the selection of
2000;115(2):205–30. the best rule in a genetic fuzzy learning algorithm. Fuzzy Sets Syst
[43] Abu Eldahab YE, Saad NH, Zekry A. Enhancing the maximum power point 2001;120(2):309–21.
tracking techniques for photovoltaic systems. Renew Sustain Energy Rev [73] Pakhira MK, Bandyopadhyay S, Maulik U. A study of some fuzzy cluster validity
2014;40:505–14. indices, genetic clustering and application to pixel classification. Fuzzy Sets Syst
[44] Piegari L, Rizzo R. Adaptive perturb and observe algorithm for photovoltaic 2005;155(2):191–214.
maximum power point tracking, renewable power generation. IET [74] Maulik U, Bandyopadhyay S. Fuzzy partitioning using a real-coded variable-length
2010;4:317–28. genetic algorithm for pixel classification. IEEE Trans Geosci Remote Sens
[45] Harrag A, Messalti S. Variable step size modified P & O MPPT algorithm using GA- 2003;41(5):1075–81.
based hybrid offline/online PID controller. Renew Sustain Energy Rev [75] Kim D, Kim C. Forecasting time series with genetic fuzzy predictor ensemble.
2015;49:1247–60. IEEE Trans Fuzzy Syst 1997;5(4):523–35.
[46] Mellit A, Kalogirou SA. MPPT-based artificial intelligence techniques for photo- [76] Alcalá-Fdez J, Alcalá R, Gacto MJ, Herrera F. Learning the membership function
voltaic systems and its implementation into field programmable gate array chips: contexts for mining fuzzy association rules by using genetic algorithms. Fuzzy Sets
review of current status and future perspectives. Energy 2014;70:1–21. Syst 2009;160(7):905–21.
[47] Ahmed EM, Shoyama M. Variable step size maximum power point tracker using a [77] Cheng PC, Peng BR, Liu YH, Cheng YS, Huang JW. Optimization of a fuzzy-logic-
single variable for stand-alone battery storage PV systems. J Power Electron control-based MPPT algorithm using the particle swarm optimization technique.
2011;11:218–27. Energies 2015;8(6):5338–60.
[48] Salas V, Olias E, Lazaro A. Review of the maximum power point tracking [78] L10-2:Khaehintung N, Kunakorn A, Sirisuk P. A novel fuzzy logic control
algorithms for stand-alone photovoltaic systems. Sol Energy Mater Sol Cells technique tuned by particle swarm optimization for maximum power point
2006;90:1555–78. tracking for a photovoltaic system using a current-mode boost converter with
[49] Scarpa V, Buso S, Spiazzi G. Low-complexity MPPT technique exploiting the PV bifurcation control. Int J Control, Autom Syst 2010;8(2):289–300.
module MPP locus characterization. IEEE Trans Ind Electron 2009;56:1513–38. [79] Hu Y, Liu J, Liu B. . A MPPT control method of PV system based on fuzzy logic and
[50] Kota VR. Renewable and Sustainable Energy Reviews; 2016, 〈http://dx.doi.org/ particle swarm optimization. In: Proceedings of IEEE international conference on
10.1016/j.rser.2016.12.054〉. intelligent system design and engineering application (ISDEA); 2012. p. 73–5.
[51] Loukriz A, Haddadi M, Messalti S. Simulation and experimental design of a new [80] Wu H, Sun Y, Meng C. Application of fuzzy controller with particle swarm
advanced variable step size incremental conductance MPPT algorithm for PV optimization algorithm to maximum power point tracking of photovoltaic
systems. ISA Trans 2016;62:30–8. generation system. Proc CSEE 2011;31(6):52–7.
[52] Bahari MI, Tarassodi P, Naeini YM, Khalilabad AK and Shirazi P. Modeling and [81] Ibrahim HEA, Elnady MA. A comparative study of PID, fuzzy, fuzzy-PID, PSO-
simulation of hill climbing MPPT algorithm for photovoltaic application. In: PID, PSO-Fuzzy, and PSO-fuzzy-PID controllers for speed control of DC motor
Proceedings of IEEE international symposium on power electronics, electrical drive. Int Rev Autom Control (IREACO) 2013;6(4):393–403.
drives, automation and motion (SPEEDAM); 2016. p. 1041–4. [82] Bingül Z, Karahan O. A fuzzy logic controller tuned with PSO for 2 DOF robot
[53] Miyatake M, Veerachary M, Toriumi F, Fujii N, Ko H. Maximum power point trajectory control. Expert Syst Appl 2011;38(1):1017–31.
tracking of multiple photovoltaic arrays: a PSO approach. IEEE Trans Aerosp [83] Bevrani H, Habibi F, Babahajyani P, Watanabe M, Mitani Y. Intelligent frequency
Electron Syst 2011;47(1):367–80. control in an AC microgrid: online PSO-based fuzzy tuning approach. IEEE Trans
[54] shaque K, Salam Z, Amjad M, Mekhilef S. An improved particle swarm Smart Grid 2012;3(4):1935–44.
optimization (PSO)–based MPPT for PV with reduced steady-state oscillation. [84] Shahzad M, Zahid S, Farooq M. July. A hybrid GA-PSO fuzzy system for user
IEEE Trans Power Electron 2012;27(8):3627–38. identification on smart phones. In: Proceedings of the 11th annual conference on
[55] Renaudineau H, Donatantonio F, Fontchastagner J, Petrone G, Spagnuolo G, genetic and evolutionary computation, ACM; 2009. p. 1617–24.
Martin JP, Pierfederici S. A PSO-based global MPPT technique for distributed PV [85] Shayeghi H, Jalili A, Shayanfar HA. Multi-stage fuzzy load frequency control using

1168
M. Nabipour et al. Renewable and Sustainable Energy Reviews 74 (2017) 1147–1169

PSO. Energy Convers Manag 2008;49(10):2570–80. Syst Appl 2009;36(4):7484–92.

[86] Wang LX. Adaptive fuzzy systems and control: design and stability analysis. Upper [95] Min YY, Liu YG. Barbalat Lemma and its application in analysis of system stability.
Saddle River, New Jersey: Prentice-Hall, Inc; 1994. J Shandong Univ (Eng Sci) 2007;1:012.
[87] Hojati M, Gazor S. Hybrid adaptive fuzzy identification and control of nonlinear [96] Wang -LX. Adaptive fuzzy systems and control: design and stability analysis.
systems. IEEE Trans Fuzzy Syst 2002;10(2):198–210. Englewood Cliffs, NJ: Prentice-Hall; 1994.
[88] Moradi M, Kazemi MH, Ershadi E. Direct adaptive fuzzy control with membership [97] Chen Chi-Tsong. Linear system theory and design. UK: Oxford University Press,
function tuning. Asian J Control 2012;14(3):726–35. Inc.; 1995.
[89] Hadjili ML, Wertz V. Takagi-Sugeno fuzzy modeling incorporating input variables [98] Mikati M, Santos M, Armenta C. Electric grid dependence on the configuration of a
selection. IEEE Trans Fuzzy Syst 2002;10(6):728–42. small-scale wind and solar power hybrid system. Renew Energy 2013;57:587–93.
[90] Angelov PP, Filev DP. An approach to online identification of Takagi-Sugeno fuzzy [99] Sharp NE-165U1, Specification sheet, 〈http://telemetryhelp.com/Datasheets/
models. IEEE Trans Syst, Man, Cybern, Part B (Cybern) 2004;34(1):484–98. sharp_165.pdf〉.
[91] Oh S, Pedrycz W. Identification of fuzzy systems by means of an auto-tuning [100] Matsui T, Sai H, Saito K, Kondo M. High-efficiency thin-film silicon solar cells
algorithm and its application to nonlinear systems. Fuzzy Sets Syst with improved light-soaking stability. Prog Photovolt: Res Appl 2013;21:1363.
2000;115(2):205–30. [101] First Solar press release, 19 March 2014. First Solar Sets Thin-Film Module
[92] Qi R, Brdys MA. Stable indirect adaptive control based on discrete-time T–S fuzzy Efficiency World Record of 17.0 Percent (〈http://investor.firstsolar.com/releases.
model. Fuzzy Sets Syst 2008;159(8):900–25. cfm?Sect=all〉; accessed 24 April 〈2014〉).
[93] Juang CF, Tsao YW. A self-evolving interval type-2 fuzzy neural network with [102] TEL Solar Press Release, July 9, 2014.
online structure and parameter learning. IEEE Trans Fuzzy Syst [103] Emery MA, Hishikawa K, Warta, W Y, Dunlop ED. Solar cell efficiency tables
2008;16(6):1411–24. (Version 45). Prog Photovolt: Res Appl 2015;23(1):1–9.
[94] Yu W, Li X. Online fuzzy modeling with structure and parameter learning. Exp

1169