Professional Documents
Culture Documents
com
ScienceDirect
Solar Energy 101 (2014) 8399
www.elsevier.com/locate/solener
Received 22 July 2013; received in revised form 28 October 2013; accepted 21 December 2013
Available online 10 January 2014
Communicated by: Associate Editor Elias K. Stefanakos
Abstract
In this paper, various intelligent methods (IMs) used in tracking the maximum power point and their possible implementation into a
recongurable eld programmable gate array (FPGA) platform are presented and compared. The investigated IMs are neural networks
(NN), fuzzy logic (FL), genetic algorithm (GA) and hybrid systems (e.g. neuro-fuzzy or ANFIS and fuzzy logic optimized by genetic
algorithm). Initially, a complete simulation of the photovoltaic system with intelligent MPP tracking controllers using MATLAB/Simulink environment is given. Secondly, the dierent steps to design and implement the controllers into the FPGA are presented, and the
best controller is tested in real-time co-simulation using FPGA Virtex 5. Finally, a comparative study has been carried out to show the
eectiveness of the developed IMs in terms of accuracy, quick response (rapidity), exibility, power consumption and simplicity of implementation. Results conrm the good tracking eciency and rapid response of the dierent IMs under variable air temperature and solar
irradiance conditions; however, the FLGA controller outperforms the other ones. Furthermore, the possibility of implementation of the
designed controllers into FPGA is demonstrated.
2013 Elsevier Ltd. All rights reserved.
Keywords: Photovoltaic system; Intelligent MPPTs; Co-simulation; Real time implementation; FPGA
1. Introduction
Tracking the maximum power point (MPP) of a photovoltaic (PV) module/array is an indispensable task of a
photovoltaic control system, since it maximizes the power
output of the PV system, and therefore maximizes the PV
module eciency. To improve the conversion eciency
of the electric power generation, a maximum power point
tracking (MPPT) algorithm is usually integrated with the
0038-092X/$ - see front matter 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.solener.2013.12.026
84
Nomenclature
STC
standard test conditions
VHSIC very-high-speed integrated circuit
Terminology
ANFIS adaptive neuro-fuzzy inference system
ANN articial neural network
ASIC application specic integrated circuit
CLBs congurable logic blocks
DSP
digital signal processor
FL
fuzzy logic
FPGA eld programmable gate array
GA
genetic algorithm
HDL hardware description language
IncCond incremental conductance
ISE
integrated software environment
MPP maximum power point
MPPT maximum power point tracking
OTP
one-time programmable
P&O perturb and observe
PV
photovoltaic
PWM pulse width modulator
RTL register transfer level
Symbols
D
duty cycle
E
error
G
irradiation (W/m2)
K
Boltzmanns constant
m
the diode ideality factor
P
power (W)
q
the charge of the electron (Cl)
T
the temperature at standard test conditions (K)
Ta
air temperature (C)
Greek
l
g
DE
DIL
Dv0
membership function
eciency
change of the error E (W)
inductor ripple current (A)
output voltage ripple amplitude (V)
Table 1
Comparison between some previous MPPT methods and MPPT-based AI techniques.
Methods
Eciency
Complexity level
Response/speed
Hardware implementation
Cost
Power consumption
P&O
IncCond
Open circuit voltage
Fuzzy logic
Neural networks
Genetic algorithm
Low
Moderate
Low
High
High
High
Simple
Simple
Simple
High
High
Moderate
Slow
Medium
Medium
Fast
Fast
Fast
Easy
Easy
Easy
Relatively easy
Dicult
Dicult
Inexpensive
Inexpensive
Inexpensive
Expensive
Expensive
Expensive
Low
Low
Low
Low
Low
Low
85
P n P n 1
V n V n 1
DEn En En 1
Load
DC-DC Buck
converter
Battery
PV module
BP MSX120Wp
Control unit
86
Fig. 3a. The employed neural network architecture for the MPPTANN
controller.
and so forth:
The basic idea is that if the last change in the duty ratio
(D) has caused the power to rise, keep moving in the same
direction; otherwise, if it has caused the power to drop
move it in the opposite direction. Fig. 2 depicts the surface
of the fuzzy logic controller. It should be noted that, fuzzy
logic methods depend on a careful selection of parameters,
denition of membership functions, and fuzzy rules.
2.2. MPPT-based neural network controller
ANNs are widely accepted as a technology oering an
alternative way to solve complex problems. The developed
neural network architecture for the MPPT controller is
given in Fig. 3a, the input variables are E, DE and the
output variable is the parameter D. These values are
Table 2
Fuzzy logic inference table (Messai et al., 2011a).
DE=E
NB
NS
ZE
PS
PB
NB
NS
ZE
PS
PB
ZE
ZE
PS
NS
NB
ZE
ZE
ZE
NS
NB
PB
PS
ZE
NS
NB
PB
PS
NB
ZE
ZE
PB
PS
NS
ZE
ZE
was chosen after several tests, which showed that the tangential sigmoid function converges faster compared to the
exponential sigmoid function during the learning phase.
The well-known LevenbergMarquardt (LM) algorithm
was used to train the network and the root mean square
error is used as criterion to stop the training process.
Fig. 3b shows the evolution of the performance error for
the designed ANN-controller. As can be observed, the
mean squared error during the training process is about
105 for ve neurons in the hidden layer. The calculated
duty cycle versus the desired one is given in Fig. 3c and
as it can be seen a good agreement between ANN-predicted
and calculated D is observed; the relative error is about 1%.
2.3. MPPT-based ANFIS controller
Combining fuzzy logic and neural networks is a powerful tool in control, prediction and modelling of complex
systems such as photovoltaic systems. MPP tracking using
87
w1 D1 w2 D2 w25 D25
w1 w2 w25
0.5
D desired
0.45
D calculated
0.4
0.35
0.3
0.25
0.2
0.15
0.1
0
50
100
150
Time (S)
88
Table 3
ANFIS inference table (Chekired et al., 2012).
DE=E
B1
B2
B3
B4
B5
A1
A2
A3
A4
A5
D1
D6
D11
D16
D21
D2
D7
D12
D17
D22
D3
D8
D13
D18
D23
D4
D9
D14
D19
D24
D5
D10
D15
D20
D25
FLC, Gas are used to nd the optimal membership functions and parameters as reported in Larbes et al. (2009)
and Messai et al. (2011b). This is achieved by implementing
the following stages.
2.4.1. The optimization criterion
The quadratic criterion to be minimized is Larbes et al.
(2009):
Z
2
J P max P dt
5
where P the desired power and Pmax the maximum power
delivered by the module under the STC (T = 25 C and
G = 1000 W/m2).The aim of this choice is to improve the
response time and reduce the uctuations.
2.4.2. Creation of the initial population
In the design of the proposed optimal FLC, two inputs
(E, DE) and one output (D) are used. Each variable is
Value
Representation
Population size
Generations
Number of genes Ng
Rate of crossover
Binary
100
50
12
100%
Mutation method
Reintegration method
Rate of mutation
Distribution index
Continuous variables
Elitist
1/12 = 8.33%
n=5
89
Rs
Id
Rsh
Iph
Vt
1
I I Ph I 0 e
6
Rsh
where Iph is the photo-generated current in STC, I0 is the
dark saturation current in STC, Rs is the panel series resistance, Rsh is the panel parallel (shunt) resistance, Vt is the
thermal voltage, given by:
Vt
mKT STC
q
but the loop closes through the now forward biased diode
(Fig. 6c). Controlling switch position the output voltage
can be maintained at a desired level lower than the input
source voltage. Thus, the buck converter can be described
by the following set of equations:
diL mL V i V O RL I L
dt
L
L
dV c1 iC1 I i I L
dt
C1
C1
and
8
dV C2 iC2 I L I O
dt
C2
C2
V i V O RL I L
DTs
L
Fig. 6a. Matlab/Simulink based simulation of the designed MPPTs for a stand-alone photovoltaic system.
10
90
V icc
Ii IL
DTs and
C1
V occ
IL Io
DTs
C2
11
where ILcc is the value of the inductor current, Vicc and Vocc
are the values of capacitors voltages. While the output voltage ripple amplitude Dmo usually ranges within 1% of the
DC component Vo, the amplitude of the inductor current
ripple DIL I Lmax I Lmin varies by as much as 1020% of
its DC value IL (Erickson, 1997). The values of L, C1
and C2 can be calculated as:
Vi Vo RL I L
DTs
2DI L
Ii IL
IL Io
C1
DTs and C 2
DTs
2DVi
2DV o
L
12
13
f s
14
Rbs
Cbp
Rbp
Voc
Rb1
Vb
91
120
100
80
60
40
20
0
10
12
14
16
18
20
Dudy cycle D
0.5
0.4
Neural networks
0.3
ANFIS
0.2
Fuzzy optimized by GA
Fuzzy logic
0.1
0
10
12
14
16
18
20
Time (S)
Fig. 7a. PV power evolution of the simulated MPPT and its corresponding duty cycle D during a test period (from 0 to 20 s) at STC.
Table 5
Eciency, response time and duty cycle rate of each technique in STC.
MPPT-controller
Eciency
g (%)
Response
time (s)
Duty cycle
rate D (%)
FL
FLGA
ANFIS
ANN
98.12
98.72
98.33
98.43
12.03
3.03
15.88
11.96
58.11
59.8
59.44
59.55
92
1000
800
600
0
10
15
20
25
30
35
20
25
30
35
20
25
30
35
120
100
80
Neural networks
60
Fuzzy logic
40
Fuzzy optimized by GA
ANFIS
20
0
0
10
15
10
15
0.6
Dudy cycle D
0.5
0.4
0.3
0.2
0.1
0
Time (S)
Fig. 7b. PV power evolution of the simulated MPPT and its corresponding duty cycle D in rapid variation of irradiance (from 1000 W/m2 to 500 W/m2
during a transition time of 5 s), air temperature was kept at 25 C.
Table 6
Eciency, response time and duty cycle rate of each technique in rapidly
changing conditions of irradiance.
MPPT-controller
Eciency
g (%)
Response
time (s)
Duty cycle
rate D (%)
FL
FLGA
ANFIS
ANN
96.4
97
96.25
96.66
8.15
3.12
7.92
7.84
42.98
43.07
42.32
42.87
T (K)
93
320
310
300
0
10
15
20
25
30
35
120
100
80
Neural networks
60
Fuzzy logic
40
Fuzzy optimized by GA
ANFIS
20
0
10
15
10
15
20
25
30
35
20
25
30
35
Dudy cycle D
0.6
0.4
0.2
Time (S)
Fig. 7c. PV power evolution of the simulated MPPT and its corresponding duty cycle D in rapid variation of air temperature (from 45 C to 20 C during
a transition time of 5 s), solar irradiance was kept at 1000 W/m2.
Table 7
Eciency, response time and duty cycle rate of each technique in rapidly
changing conditions of air temperature.
MPPT-controller
Eciency
g (%)
Response
time (s)
Duty cycle
rate D (%)
FL
FLGA
ANFIS
ANN
98.07
98.20
97.52
97.86
8.23
2.97
7.87
7.75
58.8
58.67
58.9
58.93
Synthesis
Synthesis involves compiling the VHDL code with
tools (e.g. Xilinx Foundation ISE 11.1i) which is a commercially available tool. The result of this compilation is
a ip-op and logic function transcription of the
high-level functionalities. Some functions can be resolved
in dierent ways, depending on the target component.
VHDL codes can be simulated using ModelSim Xilinx
or other tools.
Placement-and-Routing
The result of the Placement-and-Routing is the nal
code to be implanted on the FPGA. An auxiliary result is
the VHDL le giving the operation of the implanted code
and taking the propagation times of the target device into
account. This le can be used in co-simulation and this
result in a representation of a virtual prototype. This allows
checking that the Placement-and-Routing has not
altered the performance and that the synchronization of
all signals is compatible with the propagation times.
4.3. The designed VHDL-modules
The controllers have been designed using the VHDL,
integrated with the Xilinx foundation ISE11.1i tools
(Xilinx, Inc.). The dierent designed VHDL-modules can
be summarized to main and secondary blocks, as:
94
MPPT algorithm
Frequency divider
95
Fig. 9b. Floor planning of the FPGA implementing the MPPT (FLGA)
controller and interface circuits (routing circuit).
observed that the used Virtex 5 is largely sucient to implement the FLGA controller. With regard to the memory
space (area) and power consumption required by each
intelligent controller, Table 8 reports the FPGA logic
resources used to develop each controller. It is well known
that power consumption is strongly dependent on the target circuit including resource utilization, low-level features
such as logic partition, mapping, placement and route. As
can be seen from Table 8 the consumption power for each
model is less than 1 W. X power analyzer available in ISE
tool is used to evaluate the power consumption. The total
power consists of the sum of dynamic and static power;
the dynamic is due to the component activity while static
power represents the power consumed by the leakage current. ChipScope Pro Analyser of ISE could be also used
to analyse the designed controller.
Fig. 9a. The evolution of the PV power and duty cycle vs. time of the designed FLGA for rapid variation of solar irradiation and air temperature
(G = 700 W/m2, 24 C ! G = 300 W/m2, Ta = 19 C ! G = 700 W/m2, Ta = 24 C).
96
Table 8
Device utilization summary and power consumption for each MPPT controller.
Device utilization summary Virtex-5
(XC5VLX50)
Fuzzy logic
controller
Neural
controller
ANFIS
controller
Available
3448 (11%)
164 (1%)
3453 (11%)
18 (4%)
5029 (17%)
0 (0%)
3381(11%)
18 (4%)
457 (1%)
107 (1%)
457 (1%)
18 (4%)
3773 (13%)
224 (1%)
3756 (13%)
18 (4%)
28,800
28,800
28,800
440
2 (6%)
2 (4%)
2 (6%)
2 (4%)
2 (6%)
47 (98%)
4 (12%)
1 (2%)
32
48
0.04146
0.42532
0.46678
52.1
0.00072
0.42459
0.42531
51.9
0.03738
0.42528
0.46266
52.1
0.04260
0.42538
0.46798
52.1
5
4
Current (A)
G=700 W/m
Tc=24 C
(a)
3
2
1
Laboratoryfacilities
Jijel Univ.
10
12
14
16
18
20
Voltage (V)
BP Solar MSX 120W
2
1.8
1.6
G=300 W/m
Tc=19 C
Current (A)
1.4
(b)
1.2
1
0.8
0.6
0.4
0.2
0
10
12
14
16
18
Voltage (V)
Fig. 10. Laboratory facilities (a) measured IV characteristic (G = 700 W/m2, Ta = 24 C) and (b) measured IV characteristic (G = 300 W/m2,
T = 19 C).
97
Fig. 11. (a) The FPGA chip used for co-simulation, and (b) the duty cycle D generated by the FLGA controller, displayed in PWM form under constant
conditions (G = 700 W/m2, Ta = 24 C) using real measured IV curve.
Table 9
Comparison of dierent intelligent MPPT controllers according to their complexity level, rapidity, eciency and space memory required in the FPGA.
Controllers
Complexity level
Eciency
FLGA
FL
ANN
ANFIS
Complex
Simple
Medium
Medium
Very fast
Relatively fast
Fast
Relatively fast
More ecient
Ecient
Ecient
Ecient
Negligible
Low
Low
Low
45
45
25
12
98
The ANFIS controller exhibits good results compared to the ANN-controller for the point of view
of time response, eciency and required space memory. However, its robustness depends on the tuning
of FL parameters and it is relatively easy to be
implemented compared to FLGA. It performs also
better in rapid variation of weather conditions.
The fuzzy controller also can provide acceptable
results and require large memory space and easy
implementation compared to other techniques.
However, the shortcoming of fuzzy computation is
obtaining the correct fuzzy rules and membership
functions, which heavily rely on the prior knowledge
of the system and has an impact on the robustness
and reliability of the method. It could also be used
in the case of rapid variation of weather conditions.
Low power is consumed by the designed MPPT controller, which is less than 0.5 W. Space memory in
this Virtex 5 FPGA is largely sucient to implement
our controllers.
providing the materials and the computer facilities for performing the present work. This work was also supported by
the TWAS under Grants (Ref. 09-108 RG/REN/AF/
AC_C: UNESCO FR:3240231224, 12-194RG/REN/AF/
AC_C: UNESCO FR:3240270869).
Appendix A
PV module specications.
Designation
Nominal power
Voltage at MPP
Current at MPP
Short-circuit current
Open-circuit voltage
BP Solar MSX120(W)
120 W
17 V
7.06 A
7.92 A
21.2 V
References
Altera, Inc. <http://www.altera.com>.
Chao, K.H., Li, C.J., 2010. An intelligent maximum power point tracking
method based on extension theory for PV systems. Expert Syst. Appl.
37, 10501055.
Chekired, F., Larbes, C., Mellit, A., 2012. Comparative study between two
intelligent MPPT-controllers implemented on FPGA: application for
photovoltaic systems. Int. J. Sustain. Energy, 742896. http://
dx.doi.org/10.1080/14786451.
Chekired, F., Larbes, C., Rekioua, D., Haddad, F., 2011. Implementation
of a MPPT fuzzy controller for photovoltaic systems on FPGA circuit.
Energy Procedia 6, 541549.
Chettibi, N., Mellit, A., Drif, M., 2012. FPGA-based implementation of
IncCond algorithm for photovoltaic applications. In: IEEE, 24th
International Conference on Microelectronics, Algeria, December
2012, pp. 1720. doi:10.1109/ICM.2012.6471401.
Erickson, R.W., 1997. Fundamentals of Power Electronics. Chapman &
Hall, 115 Fifth Avenue, New York, USA.
Kulaksz, A.A., Akkaya, R., 2012. A genetic algorithm optimized ANNbased MPPT algorithm for a stand-alone PV system with induction
motor drive. Sol. Energy 86, 23662375.
Khaehintung, N., Wiangtong, T., Sirisuk, P., 2006. FPGA Implementation of MPPT using variable step-size P&O Algorithm for PV
Applications. In: IEEE International Symposium on Communication
and Information. http://dx.doi.org/10.1109/ISCIT.2006.340033, pp.
212215.
Koutroulis, E., Kalaitzakis, K., Tzitzilonis, V., 2009. Development of an
FPGA-based system for real-time simulation of photovoltaic modules.
Microelectron. J. 40, 10941102.
Larbes, C., AtCheikhS, M., Obeidi, T., Zerguerras, A., 2009. Genetic
algorithms optimized fuzzy logic control for the maximum power point
tracking in photovoltaic system. Renew. Energy 34, 20932100.
Liu, Y.H., Liu, C.L., Huang, J.W., Chen, J.H., 2013. Neural-network-based
maximum power point tracking methods for photovoltaic systems
operating under fast changing environments. Sol. Energy 89, 4253.
Lu, C.F., Liu, C.C., Wu, C.J., 1995. Dynamic modelling of battery energy
storage system and application to power system stability. IEE Proc.
Generation Transm. Distribution 142, 429435.
Mekki, H., Mellit, A., Kalogirou, S.A., Messai, A., Furlan, G., 2010.
FPGA-based implementation of a real time photovoltaic module
simulator. Prog. Photovolt: Res. Appl. 18, 115127.
Mellit, A., Kalogirou, S.A., 2008. Articial intelligence techniques for
photovoltaic applications: a review. Prog. Energy Combust. Sci. 34,
574632.
99
Shaiek, Y., BenSmida, M., Sakly, A., Mimouni, M.F., 2013. Comparison
between conventional methods and GA approach for maximum power
point tracking of shaded solar PV generators. Sol. Energy 90, 107122.
Salam, Z., Ahmed, J., Merugu, B.S., 2013. The application of soft
computing methods for MPPT of PV system: a technological and
status review. Appl. Energy 107, 135148.
SalamehZ, M., Casacca, M.A., Lynch, W.A., 1992. A mathematical model
for leadacid batteries. IEEE Trans. Energy Convers. 7, 9398.
Sera, D., Teodorescu, R., Rodriguez, P., 2007. PV panel model based on
datasheet values. Proc. IEEE Int. Symp. Ind. Electron. http://
dx.doi.org/10.1109/ISIE.2007.4374981, pp. 23922396.
Subudhi, B., Pradhan, R., 2013. A comparative study on maximum power
point tracking techniques for photovoltaic power systems. IEEE
Trans. Sustain. Energy 4, 8998.
Xilinx Corporation, Inc. <http://www.xlinx.com>.
Xilinx, System Generator for DSP User Guide, Release 10.1, March 2008.
Youssef, E.B., Stephane, P., Bruno, E., Corinne, A., 2010. New P&O
MPPT algorithm for FPGA implementation. In: IECON 2010, 36th
Annual Conference on IEEE Industrial Electronics Society, vol. 710,
November 2010, pp. 28682873. doi:10.1109/IECON.2010,5675079.