2017 Nineteenth International Middle East Power Systems Conference (MEPCON), Menoufia University, Egypt, 19-21 December 2017

New Maximum Power Point Tracking Technique of a

Grid-Connected PV System with Subtractive Clustering
Diaa M. Gawad Mostafa A. Elshahed, M. A. Moustafa Hassan
National Energy Control Center, Electrical Power and Machines Dept.,
Ministry of Electricity and Renewable Energy, Faculty of Engineering, Cairo University,
Cairo, Egypt Cairo, Egypt

Abstract— With great increasing concerns about fossil fuel composes PV system. It contains solar panels, Maximum
shortage, global warming, and damage to the environmental Power Point Tracking (MPPT), solar inverter besides storage
system, the encouraging to increase in the alternative energy batteries [2]. The solar cell model is fabricated from
resources which have low operation cost and low pollutant semiconductor materials commonly silicon. An electrical one-
emissions have a great concern. The photovoltaic (PV) system is diode model representation defines the solar cell model [3].
one of the renewable energy resources widely spread in the The MPPT is a simple DC-DC converter which is inserted
energy market in the globe as well as Egypt. Several natural between the PV array and the load. Its control is accomplished
conditions affect the output power of the PV system such as over different tracking algorithms.
solar irradiance and ambient temperature. Maximum Power
Point Tracking (MPPT) techniques for controlling DC-DC boost A solar tracking system may be used to improve the
converter tend to reach for the maximum output power. Various overall performance of the system [4]. A large number of
MPPT based techniques have been designated in the literature. MPPT techniques are introduced over the last decades such as
In this paper, the Adaptive Neuro-Fuzzy Inference System Perturbation and Observation (P&O), Fuzzy Logic Controller
technique based on Subtractive Clustering method is proposed (FLC), Adaptive Neuro-Fuzzy Inference System (ANFIS)
and compared with other previously used techniques. A grid- based on Grid Partition generating Fuzzy Inference System
connected PV system of 100kW is considered and investigated. (FIS) [5] in addition to ANFIS based on Subtractive
From results, the system response with the proposed technique
Clustering (Sub-Clust) generating FIS. P&O algorithm is the
for MPPT is faster, robust, and more stable than other
most commonly used technique in practice because of its ease
techniques. In addition, there are no any overshoots in the DC
link voltage and the value of AC output power is increased the
of implementation but it has unavoidable oscillation problem
obtained results are promising. [6]. FLC is a very successful method in the application for
finding maximum power point but it is a non-linear control
Index Terms-- Adaptive Neuro-Fuzzy Inference System; Grid- method as presented in [7]. To design a compatible fuzzy-
Connected PV System; Maximum Power Point Tracking; based controller it is necessary to make use of the expert
Subtractive Clustering Method. knowledge of an experienced user. Under variations of the
environmental parameters, the controller provides robust
I. INTRODUCTION performance [7].
The continuous increase in energy demand, with nation’s Neural networks may be combined with fuzzy logic
economic development, pushes looking for alternative sources resulting in a new robust Artificial Intelligence (AI) method
of energy (renewable energy) that is not harmful and does not so-called ANFIS. The information extracted from the fuzzy
damage the environment system. Among the used renewable control experiences is used to initialize the parameters of
energy alternatives, the energy through the Photovoltaics ANFIS controller. The parameters of ANFIS are updating
(PVs) is one of the widely spread energy sources. It can be adaptively by using an online learning algorithm based on the
considered the most necessary and qualified sustainable Back-Propagation (BP) method. ANFIS is able of coming
resource. In spite of the non- continuous sunlight, solar energy with the input-output mapping of training data sets when
is commonly available and totally free of cost. Lately, PV trained with a sufficient number of epochs [8].
panels are familiar and can be widely utilized to the forefront In this paper, a new MPPT technique with using ANFIS
in electric power applications. These panels have the ability to based on Sub-Clust method is proposed and compared with
provide DC electricity without environmental influence and previous techniques in the literature. The MPPT techniques
contamination free when is exposed to solar radiation [1]. for the DC-DC boost converter based on Perturbation and
The solar panel is the vital part of a complete PV system. Observation, Fuzzy Logic Controller, and ANFIS technique
It is considered a power system that supplies solar power by based on Grid Partition method are to be compared with the
means of PV. An arrangement of several components proposed ANFIS technique based on Sub-Clust method. In

978-1-5386-0990-3/17/$31.00 ©2017 IEEE

888
addition to the introduction in this section, the paper is MPPT approach is to periodically differ the next step direction
organized as follows: a brief description of the implemented by a constant factor ± ΔPpv ΔV pv [4].
grid-connected PV system is discussed in section 2,
clarification of various MPPT techniques is illustrated in Anyway, the tracking point firstly begins; the final
section 3, then simulation results achieved are presented in objective is to reach the steady state operation area that will be
section 4, and finally the conclusion is given in section 5. nearby maximum power point [8]. The advantages of this
algorithm, as listed previously [7], are simplicity and easiness
II. GRID-CONNECTED PV SYSTEM of execution. In any case, P&O has constraints that decrease
The grid-connected PV system, as shown in Fig. 1, its MPPT effectiveness as discussed in [9]. One such
consists of the PV modules connected to the grid through a constraint is that when the amount of sunlight declines, the P–
DC-DC converter controlled by various ways of MPPT V curve levels out [11].
techniques and an inverter. This leads to difficulty for the MPPT to discover the
location of the maximum power point, causing a small change
in power concerning the perturbation of the voltage. Another
major disadvantage of P&O is that it oscillates nearby the
maximum power point, varying the sign of the perturbation
after any estimation of ¨P. Similarly, it has been demonstrated
that P&O can exhibit unreliable performance under rapidly
changing irradiance levels [12].
Fuzzy Logic Controllers (FLC) Based MPPT give the
benefits of treating with imprecise inputs, not requiring an
exact mathematical model, besides it can handle nonlinearity
[13]. The structure of fuzzy logic controller system composed
Figure 1. Grid-connected PV system components. of four essential components which are fuzzifier, rule base,
inference engine and defuzzifier. A simple configuration of
The PV modules model, mentioned above, includes a the fuzzy logic controller system is introduced Fig. 2 and
current source I ph , single diode as well as a series resistance displayed in [14] recently. In Fig. 2 E(k) and CE(k) are Error
and Change of Error respectively at an instant (k).
RS as displayed in [3]. This series resistance combines both
the resistance inside every cell and the resistance in the
connection between the cells. The total output current is the TABLE I. PARAMETERS OF PV MODEL 100-KW330 SUNPOWER
(SPR-305).
difference between the photocurrent I ph and the diode
current I D : Parameter Value
Voc 64.2 V
I = I ph − I D = I ph − I o ª¬exp ( qV mkT ) − 1º¼ (1) Isc 5.96 A
Vmp 54.7 V
Where;
m is idealizing factor, Imp 5.58 A
k stands for Boltzmann’s gas constant, Nser 5 Panels
T is the absolute temperature of the cell in °k, Npar 66 Panels
q represents an electronic charge, K 1.38 x 10^-23 J/K
V is the voltage imposed on the cell,
Q 1.6 x 10^-19 C
I0 is the dark saturation current, which is also identified as
the leakage or diffusion current which is mainly T 25 oC
depending on temperature [9]. Ir 1000 W/m2

The solar module parameters are presented in Table 1, as

adopted from [10]. The MATLAB simulation based on the
datasheet of SunPower (SPR-305) PV modules [6].

III. MPPT CONTROL TECHNIQUES

Several MPPT techniques have been carefully considered
and well developed to produce sufficient duty cycle signal for
proposed MPPT technique. Three different techniques are
chosen to be discussed and compared with the proposed
ANFIS technique based on subtractive clustering (Sub-Clust)
method. The principle of Perturbation and Observation (P&O)

Figure 2. The basic structure of FLC system.

889
 From the simulated PV model, there is a group of data set
used for training ANFIS. It can be collected by recording
Initially, fuzzification step, a crisp set of input data (E(k) each pair of the input data (DC Link Voltage and current of
and CE(k)) are assembled and then transformed into a fuzzy PV) and duty cycle as the output of the controller. This input-
set by using fuzzy linguistic variables, fuzzy linguistic output table is in form of crisp data. ANFIS has the skillful of
relationships, and membership functions. Finally, an developing the input-output mapping of these data sets once it
inference is prepared based on a group of rules (stored in the is trained with an adequate number of epochs [18].
Knowledge Base). Defuzzification step, the succeeding
output of fuzzy is transformed again to a crisp output using ANFIS produces the set of fuzzy rules to develop
the membership functions [15]. appropriate output for different values of inputs. Parameters
of membership functions are adjusted or improved so that the
The rule base, in a fuzzy system, is designed to adjust the error is minimized to the least value. When totally the
output variable (Duty cycle). A fuzzy rule can be expressed parameters of the membership function are adjusted, then
by a simple IF-THEN rule using a condition and conclusion. ANFIS model becomes organized to control the MPPT
It can be described by the matrix table as illustrated in Table scheme. Though, before using the ANFIS learning model for
1. The basic input parameters Error (E) and Change in Error the MPPT control, the results are checked by using the
(¨E) are placed along the axes, and the conclusions are laid checking data that is different from the training data. If the
on the table. A 49 rule base is implemented for the grid- error is more than the acceptable value, parameters of
connected PV system to achieve maximum power with high membership functions are adjusted to eliminate the error [19].
accuracy.
Fig. 3 represents the flow chart of ANFIS technique
The system contains two inputs (I_PV, V_PV) and one based MPPT. The data can be collected and trained by
output (D). The inputs are described generally by an error ANFIS editor [20]. To train the ANFIS a group of 10000
E (k ) and a change in error ΔE (k ) , as described in equations training data sets with 333 epochs used and training error can
(2), (3): be reduced to about 1%. ANFIS, therefore, creates the Fuzzy
Inference System (FIS) file by the group input-output data
ΔI I sets and the membership function parameters. MFs are altered
E (k ) = + (2)
ΔV V by means of the hybrid optimization process that is an
arrangement of the least-squares type and back propagation
ΔE (k ) = E (k ) − E (k − 1) (3) algorithm [21].
Where; I is the DC output current from PV module, ΔI
equals I (k ) − I (k − 1) , V is the DC output voltage from PV
module and ¨V equals V (k ) − V (k − 1) .
Recently, Adaptive Neuro-Fuzzy Inference System
(ANFIS) technique has been recommended for MPPT in the
PV systems. An adaptive neural network has the benefits of
learning capability, optimization, and balancing. Though, a
fuzzy logic is a technique based on rules created by the expert
knowledge of the designers [16]. The great performance and
efficiency of FLC have been appropriate in nonlinear and
complex systems.
ANFIS associates the benefits of using a mix of adaptive
neural network and fuzzy logic together. ANFIS creates an
input-output mapping that based both on knowledge of the
experts (in the form of fuzzy if-then rules) and the generated
input-output data resulting from using a hybrid algorithm that
is the grouping of the least-squares and backpropagation
gradient descent method [17].

TABLE II. RULE BASE OF FUZZY LOGIC CONTROLLER.

Figure 3. Flowchart of ANFIS based MPPT

890
 The ANFIS structure implemented by the MATLAB is
shown in Fig. 4. It includes five layers. The two inputs
represent DC link voltage and current output of PV model.
Both the two inputs have 7 membership functions. The
structure displays the two inputs which are expressed into
proper membership functions. This network is trained for 333
epochs. The training process is ended as soon as the outputs
of the trained system reach very close to the proposed outputs
[22]. There are two methods used by ANFIS to generate FIS:
i) Grid Partition
In this method the number also the type of input and
output of membership functions (MFs) are selected manually
[5], [15].
ii) Subtractive Clustering
This method is used when the number of clusters there
should be for a given set of data are not obviously
recognized. Sub-Clust contributes the benefit that it has fast
response and has one-pass algorithm to estimate the number Figure 4. Implemented ANFIS model structure.
of clusters and the cluster centers in a group of data. The
estimated cluster, achieved from the subclust function, is used Sub-Clust method benefits from the MATLAB subclust
to initialize iterative optimization-based Sub-Clust function to develop the clusters. This method is proposed in
approaches and model identification methods (like ANFIS). the current research and compared with other controllers to
prove its effectiveness.

Figure 5. PV system connected to grid MATLAB model.

IV. SIMULATION AND RESULTS: These all controllers are implemented using MATLAB
/Simulink applying to a 100kW grid-connected PV system as
displayed in Fig. 5. This model is put up originally by a team
The comparison results of P&O, FLC, ANFIS technique of math works then adapted to make use of each part to get the
based on Grid Partition and ANFIS technique based on Sub- best performance grid-connected PV system [6]. The
Clust based MPPT is investigated in this section. The simulation will be performed for the above methods using the
considered four MPPT Techniques are listed as follows: same input and the same load and grid.
• Perturbation and Observation P&O [6]. The performance of PV array is affected by many
• Fuzzy logic Controller [7]. influences. Irradiance and ambient temperature are essential
• ANFIS technique based on Grid Partition generating FIS factors that disturb the output of PV model. They can be
[5], [15]. displayed in Fig.s 6 and 7 as inputs of the PV system. In the
• The proposed ANFIS technique based on Sub-Clust Fig.s 6 and 7, the collected data for the sun is drawn as curves
generating FIS. that present irradiance and temperature, respectively.

891
 Figure 6. Irradiance (Ir) as the input of PV arrays.

Figure 8. DC terminal voltage of PV arrays in response to the variations in

Ir and T with using the different MPPT techniques.

Figure 7. Temperature (T) as the input of PV arrays.

It is very important to consider a wide range of irradiance

and temperature to show the performance of the PV model.
The temperature effect on Voc is negative that means the
voltage will reduce when the temperature increases as clearly
shown in Equation (4). Figure 9. The duty cycle of the different MPPT techniques.

STC KV , %
VOC (T ) = VOC + (T − 273.15) (4)
100 The value of duty cycle is in the range from 0.38 to 0.58,
Then Fig.s 8~13 show the DC Terminal Voltage of PV and it is very good to track maximum output power by
Arrays, duty cycle, DC Link Voltage and PV Output AC adjusting PV output voltage. By using P&O algorithm the
Power in response to the variations in irradiance and value of duty cycle looks like a nonlinear performance
temperature by implementing the different MPPT techniques. whereas the others look good. The duty cycle generated by
For the different previously discussed controllers, it is noted ANFIS technique based on Sub-Clust method achieve a
that the output power is affected by irradiance and preferable performance and consequently the value of output
temperature. The increasing or decreasing of the value of this voltage and power as shown.
input leads to power minimization. The proposed output Fig. 10 introduces DC Link Voltage of PV model after
power is achieved at 25c° and 1000w/m2 as stated at regulation using the four different values of duty cycle
Standard Test Condition (STC) [3]. produced. To distinguish between the DC links voltages of
Fig. 8 illustrates the DC terminal voltage of PV arrays each controller, Fig. 10 is magnified in Fig. 11. The duty
according to input parameter variations. It generally depends cycle variation boosts the DC terminal voltage from 273.5V
on temperature. The current rises with temperature but very to around 500V. It can be seen from this figure that the
little and it does not regulate the reduction in the voltage voltages using the four considered techniques have major
produced by temperature growth and consequently power differences. In the P&O technique, there are high overshoots
moreover decreases. The DC terminal voltage and current are in the voltage response.
used as inputs for MPPT controller to detect power as well as The ANFIS technique based on Sub-Clust method appears
generating a reference voltage (Vref) to track the maximum as more stable and faster than the others, which can be useful
power point. Fig. 9 illustrates duty cycle of the different four when connected to load through an inverter.
techniques considered. This signal is fed to boost converter to
raise the voltage. At the beginning of simulation (0:0.05sec),
the controller is disabled and there is no pulse to boost
converter also the system appears to be unstable [5].

892
 Figure 10. DC link voltage in response to the variations in Ir and T with Figure 12. PV output AC power in response to the variations in Ir and T with
using the different MPPT techniques. using the different MPPT techniques.

Figure 13. Magnified PV output AC power in response to the variations in Ir

and T with using the different MPPT techniques.

Figure 11. Magnified DC link voltage in response to the variations in Ir and

T with using the different MPPT techniques.

Fig.s 12 and 13 introduce AC output power that fed to the V. CONCLUSION

load and the grid after the voltage source inverter and its The grid-connected PV system comprises of PV modules,
magnification, respectively. Figures show how the different DC-DC boost converter as well as voltage source inverter tied
values of irradiance and ambient temperature affect the to the grid. The maximum power point tracking (MPPT)
amount of generated power from the PV array when using techniques for the DC-DC boost converter based on
different MPPT techniques. It is noticed that during the time Perturbation and Observation (P&O), Fuzzy logic controllers
from 4 sec to 5 sec, the AC output power is increased due to (FLC), Adaptive Neuro-Fuzzy Inference System (ANFIS)
temperature variation with constant irradiance. During the with Grid Partition method and proposed ANFIS technique
time from 5 sec to 6 sec, it can be seen that P&O technique based on the Sub-Clust method are investigated. In this paper,
introduces the highest output AC power at only low a 100kW grid-connect PV system is simulated under the effect
temperatures. of varying irradiance and temperature with using the proposed
ANFIS technique based on Sub-Clust and the results are
As realized from the results, the AC output power of PV is compared with the cases using different MPPT techniques in
directly proportional to the irradiance. The power output of the the literature to obtain the most efficient controller. In the case
PV model depends on the number of series and parallel cells of using the proposed technique, ANFIS technique based on
in PV array, also the type of semiconductor used in cell Sub-Clust, the system response is faster, robust, and more
fabrication [18]. The performance of the P&O algorithm is not stable than that with the other techniques. In addition, there
accurate and the response to variations in irradiance is very are no any overshoots in the DC link voltage and the value of
slow, in addition to, the high overshoots in the AC power AC output power is increased. These results proved that the
response. The ANFIS technique based on Sub-Clust method proposed ANFIS technique based on Sub-Clust method is
appears overall as more stable and faster than other outperformed among the other three techniques discussed.
techniques.

893
 REFERENCES photovoltaic systems,” IEEE Transactions on Industrial Electronics,
2009, vol. 56, no. 11, pp. 4473-4482.
[1] Huan-Liang Tsai, Ci-Siang Tu, and Yi-Jie Su. “Development of
Generalized Photovoltaic Model Using MATLAB/SIMULINK,” [13] M. M. Ismail, and M. M. Hassan, “The state of charge estimation for
Proceedings of the World Congress on Engineering and Computer rechargeable batteries based on artificial neural network techniques,”
IEEE International Conference on Control, Decision and Information
Science 2008 WCECS, San Francisco, USA, October 2008, pp.22 – 24.
[2] O. Wasynczuk, “Dynamic behavior of a class of photovoltaic power Technologies (CoDIT), 2013, Hammamet, Tunisia, pp. 733-739.
systems,” IEEE Transactions on Power Apparatus and Systems, vol. [14] N. Diaz, J. Hernández and O. Duarte “Fuzzy maximum power point
PAS-102, no. 9, 1983, pp. 3031-3037. tracking techniques applied to a grid-connected photovoltaic system,”
[3] A. Hansen, P. Lars, H. Hansen, and H. Bindner, “Models for a Stand- INTECH Open Access Publisher, 2011.
Alone PV System,” Risoe National Laboratory, Roskilde, 2000. [15] G. Balasubramanian and S. Singaravelu “Fuzzy logic controller for the
Available [Online]: http://www.risoe.Dk/rispubl/VEA/ris-r-1219.htm. maximum power point tracking in photovoltaic system,” International
[4] K. H. Hussein, I. Muta, T. Hshino, and M. Osakada, “Maximum Journal of Computer Applications, 2012, vol. 41, no.12, pp. 22–28.
photovoltaic power tracking: an algorithm for rapidly changing [16] R. Rahmani, M. S. mahmoudian, S. Mekhilef and R.Yusof,
atmospheric conditions,” Proc. Inst. Elect. Eng, Jan. 1995, vol. 142, no. “Implementation of Fuzzy Logic Maximum PowerPoint Tracking
1, pp. 59-64. Controller for Photovoltaic System,” American Journal of Applied
[5] D. Mlakiü, and S. Nikolovski, “ANFIS as a method for determination Sciences, 2013, vol. 10, no. 3, pp. 209-218.
MPPT in the photovoltaic system simulated in MATLAB/Simulink,” [17] F. M. Bendary, E. M. Saied, W. A. Mohamed, and Z. E. Afifi,
IEEE International Conference on Information and Communication “Optimal Maximum Power Point Tracking of PV Systems based
Technology, Electronics and Microelectronics (MIPRO), 2016, Opatija, Genetic-ANFIS Hybrid Algorithm,” International Journal of Scientific
Croatia, pp. 1082-1086. & Engineering Research, 2016, vol. 7, no. 4, pp. 830-836.
[18] R. Kharb, M. D. Ansari, and S. Shimi, “Design and implementation of
[6] https://www.mathworks.com/help/physmod/sps/examples/average-
model-of-a-100-kw-grid-connected-pv-array.html ANFIS based MPPT scheme with open loop boost converter for solar
[7] Abdelaziz, Almoataz, Hadi M. El-Helw, and B. Abdelhamed, PV module”, International Journal of Advanced Research in Electrical,
Electronics and Instrumentation Engineering, 2014, vol. 3, no. 1, pp.
“Transient Analysis of Grid-Connected Photovoltaic System Based on
Comparative Study of Maximum Power Point Tracking Techniques,” 6517-6524.
International Journal of Advances in Power Systems (IJAPS), 2013, [19] M. A. Aziz, M. M. Hassan, and E. A. Zahab, “Applications of ANFIS
in high impedance faults detection and classification in distribution
vol. 1, no. 3, pp. 1-8.
[8] M. A. Eltawil, and Z. Zhao, “Grid-connected photovoltaic power networks”, IEEE International Symposium on Diagnostics for Electric
systems: Technical and potential problems-A review,” Renewable and Machines, Power Electronics & Drives (SDEMPED), 2011, Bologna,
Italy, pp. 612-619.
Sustainable Energy Reviews, 2010, vol. 14, no. 1, pp. 112-129.
[20] F. Z. Amatoul, M. T. Lamchich, and A. Outzourhit, “Design control of
[9] H. Kaushik, & M. Rohal, “Performance analysis of a Grid Connected DC/AC converter for a grid connected PV systems using
Photovoltaic Power Plant with MPPT facility,” International Journal of MATLAB/Simulink”, IEEE conference on Power Electronics and
Enhanced Research in Science Technology & Engineering, 2015, vol. Applications, 2011, Birmingham, UK, pp. 1-7.
4, no. 2, pp. 89-97. [21] B. Tarek, D. Said, and M. E. H. Benbouzid, “Maximum power point
[10] A. B. Or, & J. Appelbaum, “Dependence of multi-junction solar cells tracking control for photovoltaic system using adaptive neuro-fuzzy,”
parameters on concentration and temperature,” Solar Energy Materials International Conference and Exhibition on Ecological Vehicles and
and Solar Cells, 2014, vol. 130, pp. 234-240. Renewable Energies (EVER), 2013, Monte Carlo, Monaco, pp. 1-7.
[11] N. Femia, G. Petrone, G. Spagnuolo, and M. Vitelli “Optimizing duty- [22] M. Buragohain, “Adaptive network based fuzzy inference system
cycle perturbation of P&O MPPT technique,” In Power Electronics (ANFIS) as a tool for system identification with special emphasis on
Specialists Conference, 2004, Aachen, Germany, vol. 3, pp. 1939-1944. training data minimization,” Doctoral Thesis, Indian Institute of
[12] N. Femia, G. Petrone, G. Spagnuolo, and M. Vitelli, “A technique for Technology Guwahati, 2008.
improving P&O MPPT performances of double-stage grid-connected

894