SOP 1

SOP TRANSACTIONS ON POWER TRANSMISSION AND SMART GRIDVolume 1, Number 1, December 2014

SOP TRANSACTIONS ON POWER TRANSMISSION AND SMART GRID

Comparative Study of ANN-GA and FuzzyController for Photovoltaic System in theGrid Connected ModeAlireza Rezvani*, Maziar Izadbakhsh, Majid Gandomkar, Foad HaidariGandoman, Saeed VafaeiDepartment of Electrical Engineering, Saveh Branch, Islamic Azad University, Saveh, Iran.

*Corresponding author: [email protected]

Abstract:Photovoltaic (PV) systems have one of the highest potentials and operating ways for generatingelectrical power by converting solar irradiation directly into the electrical energy. Consequently, itis important to track the generated power of the PV system and utilize the collected solar energyoptimally. This paper proposes an integrated offline genetic algorithm (GA) and artificial neuralnetwork (ANN) to track the solar power optimally based on various operation conditions dueto the uncertain climate change. Data are optimized by GA and then these optimum valuesare used in neural network training. The obtained results show minimal error of maximumpower point (MPP), optimal voltage (Vmpp) and superior capability of the suggested methodin the maximum power point tracking (MPPT). The simulation results are presented by usingMatlab/Simulink and show that the neural networkGA controller of grid-connected mode canmeet the need of load easily and have fewer fluctuations around the maximum power point; also,this method has well regulated PV output power and it produces extra power rather than fuzzylogic method for different conditions.

Keywords:Photovoltaic, Neural Network, Genetic Algorithm, Fuzzy Logic

1. INTRODUCTION

Renewable energy sources play an important role in electricity generation. Different renewable energysources such as a wind, solar, geothermal and biomass can be applied for generation of electricity andfor meeting our daily energy needs. Photovoltaic generation is becoming increasingly important as arenewable source since it offers many advantages such as incurring no fuel costs, not being polluting,required little maintenance, and emitting no noise, among others.

Photovoltaic (PV) systems have one of the highest potentials and operating ways for generatingelectrical power by converting solar irradiation directly into the electrical energy. Although, developingphotovoltaic energy sources can reduce fossil fuel dependency, PV panels are low-energy conversionefficient [1, 2].

29


In order to control maximum output power, using MPPT system is highly recommended. The outputpower of a PV module varies as a function of the voltage and also the MPP is change by variation oftemperature and sun irradiation. A DC-to-DC converter locates among PV systems and users, whichswitching operation of this converter is performed by the MPPT [3]. In the last few decades, differentmethods are utilized in order to achieve maximum power. The most prevalent technics are perturbationand observation algorithm (P&O) [3, 4] Incremental conductance (IC) [5, 6] fuzzy logic [7, 8] and ANN[9–11].

According to above mentioned research, the benefits of perturbation and observation algorithm andincremental conductance are1- low cost implementation 2-simple algorithm. The depletion of thesemethods is vast fluctuation of output power around the MPP even under steady state illumination whichresults in the loss of available energy [12–15]. However the fast variation of weather condition affects theoutput and these technics cannot track the maximum power.

Using fuzzy logic can solve the two mentioned problem dramatically. In fact, fuzzy logic controllercan reduce the oscillations of output power around the MPPT and has faster respond than P&O and IC.Furthermore, convergence speed of this way is higher than two mentioned way. One the weak point offuzzy logic comparing to neural network is oscillations of output power around the MPP [14, 15].

Nowadays, artificial intelligence (AI) techniques have numerous applications in determining the sizeof PV systems, MPPT control and optimal structure of photovoltaic systems. In most cases, multilayerperceptron (MLP) neural networks or radial basis function network (RBFN) have been employed formodeling PV module and MPPT controller in PV systems [16, 17].

ANN based controllers are applied to forecast optimum voltages corresponding to the MPP of PVsystem for different radiations and temperatures conditions. A review on AI methods applications inrenewable energy was studied in these literatures [9, 18]. Neural networks are the best estimation fornon-linear systems and by using ANN, oscillations of output power around the MPPT and time to reachthe MPP are decreased [6].

In [19–21], GA is used for data optimization and then, the optimum values are utilized for trainingneural networks and the results show that, the GA technic has less fluctuation in comparison with theconventional methods. However, one of the major drawbacks in mentioned papers that they are notpractically connected to the grid in order to ensure the analysis of photovoltaic system performance, whichis not considered.

In this paper first, temperature and irradiance as inputs data are given to genetic algorithm and optimalvoltages (Vmpp) corresponding to the MPP are obtained then, these optimum values are used in the neuralnetwork training. Photovoltaic module is connected to the grid using a P-Q controller of grid side toexchange active and reactive power and observe system efficiency in different weather conditions.

The paper is organized as follows: In part 2 detail of PV system is described. Part 3 is discussed stepsto implement the GA and neural networks, respectively. In part 4 fuzzy logic method is presented. In part5 P-Q controller is described and in part 6 the results are presented based on current study.

2. PHOTOVOLTAIC CELL MODEL

Figure 1 shows equivalent circuit of one photovoltaic array [2], [3]. Features of PV system is describedas following Equation (1) :

30

Comparative Study of ANN-GA and Fuzzy Controller for Photovoltaic System in the Grid Connected Mode

Equations

IPV = Id + IRP + I (1)

I = IPV − I0

[exp

(V +RSI

Vthn

)−1

]− V +RsI

RP(2)

Vth =NskT

q(3)

I0 = I0,n

(Tn

T

)3

exp[

q∗Eg

n∗ k

(1Tn

− 1T

)](4)

Where, I is the output current, V is the output voltage, Ipv is the generated current under a giveninsolation, Id is diode current, IRP is the shunt leakage current, I0 is the diode reverse saturation current, nis the ideality factor (1.36) for a p-n junction, Rs is the series loss resistance (.1 Ω), and RP is the shunt lossresistance (161.34 Ω). Vth is known as the thermal voltage. q is the electron charge (1.60217646×10−19C),k is the Boltzmann constant (1.3806503×10−23 J/K), T ( in Kelvin) is the temperature of the p-n junction.Eg is the band gap energy of the semiconductor (Eg ≈ 1.1eV for the polycrystalline Si at 25oC) and I0,n isthe nominal saturation current . T is the cell temperature; Tn is cell temperature at reference conditions.Red sun 90 w is taken as the reference module for simulation and the name-plate details are given inTable 1. The array is the combination of 6 cells in series and 6 cells in parallel of the 90 w modules;hence an array generates 3.2 kW.

Table 1. Red sun 90w module

IMP (Current at maximum power) 4.94 A

VMP (Voltage at maximum power) 18.65V

PMAX (Maximum power) 90W

VOC (Open circuit voltage) 22.32

ISC (Short circuit current) 5.24

NP (Total number of parallel cells) 1

NS (Total number of series cells) 36

Figure 1. Equivalent circuit of one photovoltaic array.

31


3. MPPT ANN AND GA

3.1 The Steps of Implementing Genetic Algorithm

In order to pursue the optimum point for maximum power in any environmental condition, ANNand GA technic are used. Besides, GA is used for optimum values and then optimum values are usedfor training ANN [19–21]. The procedure employed for implementing genetic algorithm is as follows[19, 22]: 1. determining the target function 2. determining the initial population size, 3. appraisingthe population using the target function, and 4. conducting convergence test stop if convergence isprovided.The target function of GA is applied for its optimization by the following: finding the optimumX =(X1,X2,X3, · · · ,Xn) to determine the F(X) in the maximum value, where the number of design variablesare regarded as 1. X is the design variable equal to PV system current and also, F(X) is the PV systemoutput power that must be maximized [21]. To determine the target function, the power should be setbased on the PV system current (IX ). The genetic algorithm structures are presented in Table 2.

Table 2. Genetic algorithm parameters

Number of Design Variable 1

Population size 20

Crossover constant 80%

Mutation rate 10%

Maximum Generations 20

F(x) =VX ∗ IX (5)

VX = ns

(v0 −

RS

npIX +(nk(T +273)/q)Ln∗

(IPV − IX/np + I0

I0

))(6)

To determine the objective function, the power should be arranged based on the current of array (IX ):

F(X) = ns

(v0 −

RS

npIX +(nk(T +273)/q)Ln∗

(IPV − IX/np + I0

I0

))∗ IX (7)

0 < IX < ISC (8)

The current constraint should be considered too. With maximizing this function, the optimum valuesfor Vmpp and MPP will result in any particular temperature and irradiance intensity.

3.2 Combination of Proposed Neural Network with Genetic Algorithm

Neural networks are most appropriate for the approximation (modeling) of nonlinear systems. Non-linear systems can be approximated by multi-layer neural networks and these multi-layer networks havebetter result in comparison with the other algorithm [16, 18]. In this paper, feed forward neural network

32


for MPPT process control is used. The important section of this technic is that, the required data fortraining process must be obtained for each PV module and each specific location [11].

Three layers can be considered for the proposed ANN. The input variables are temperature and solarirradiance and Vmpp corresponding to MPP is output variable of the neural network as shown in Figure 2.Proposed MPPT Scheme is illustrated in Figure 3. The sum of pi controller and High frequency triangularcarrier for pwm generation is conducted toward boost converter.

Figure 2. Feed forward neural network for MPPT.

Figure 3. Proposed MPPT Scheme.

Figure 4. Inputs data of irradiation and temperature.

The output of PV system has varied during time and environmental conditions. Thus, periodic trainingof the ANN is needed. Training of the ANN is a set of 500 data as shown in Figure 4. ( irradiancebetween 0.05 to 1 watt per square meter (W/m2) and temperatures between -5oC to 55oC) and also, a setof 500 Vmpp corresponding to MPP is obtained by GA as shown in Figure 5.

33


Figure 5. The output of Vmpp-Mpp optimized by (GA).

In order to implementation of the ANN for MPPT, first it should be determined the number of layers,number of neurons in each layer, transmission function in each layer and type of training network. Theproposed ANN in this paper has three layers which first and second layers have respectively 16 and 11neurons and third layer has 1 neuron. The transfer functions for first and second layers are Tansig andfor third layer is Purelin. The training function is Trainlm. The acceptable sum of squares for networkis supposed to be 10−9. Which training this neural network in 850 iterations, will converge to a desiredtarget. After training, the output of training network should be close to optimum output from GA. Figure6 show the output of the neural network training with the amount of target. A set of 80 data is used forthe ANN test. Figure 7 illustrate the output of the neural network test with the amount of target whichshowing a negligible training error percentage of about 0.3%.

4. FUZZY LOGIC CONTROLLER

Fuzzy logic is popular in the last decade. MPPT using fuzzy logic control obtains several advantages ofbetter performance, powerful and simple design. In addition, this method does not require the knowledgeof the exact model of system. It is known by multi-rules-based resolution and multivariable consideration[12]. Fuzzy logic controller is made of three parts which is demonstrated in Figure 8(a). First partis fuzzification which is the process of changing a real scalar value into a fuzzy set. Second part isfuzzy inference motor that combines IF-THEN statements based on fuzzy principle and finally, we havedefuzzification which is the process that changes a fuzzy set into a real value in output [? ]. The proposedFLC has two inputs and one output. The two FLC input variables are the error (E) and change of error(CE) as demonstrated by the following equation (9), (10). The output of FLC is duty cycle (D) that it isused, for the tracking of the MPP by comparing with the saw tooth waveform to generate a PWM signalfor the boost converter.

Equations

34


E( j) =Ppv( j)−Ppv( j−1)Vpv( j)−Vpv( j−1)

(9)

CE( j) = E( j)−E( j−1) (10)

Three triangle membership functions are considered in this article. In this paper Mamdanis fuzzyinference method, with Max-Min operation fuzzy combination is used. Given membership functionsare shown in Figure 8(b), (c) and (d). The fuzzy rule algorithm includes 25 fuzzy control rules listed inTable 3.

Table 3. FLC Rules base

ECE

NB NS ZE PS PB

NB ZE ZE PB PB PB

NS ZE ZE PS PS PS

ZE PS ZE ZE ZE NS

PS NS NS NS ZE ZE

PB NB NB NB ZE ZE

Moreover, The rules implemented to obtain the required data are shown in Table 3.The linguisticvariables are represented by ZE (zero), PB (positive big), PS (positive small), NB (negative big), NS(negative small), respectively.

5. CONTROL STRATEGY (P-Q)

Inverter control model is illustrated in Figure 9 The goal of controlling the grid side, is keeping thedc link voltage in a constant value regardless of production power magnitude. Internal control-loopwhich control the grid current and external control loop which control the voltage [23]. Also, internalcontrol-loop which is responsible for power quality such as low total harmonic distortion (THD) andimprovement of power quality and external control-loop is responsible for balancing the power. Forreactive power control, reference voltage will be set same as dc link voltage. In grid-connected mode,photovoltaic module must supply local needs to decrease power from the main grid. One the main aspectsof P-Q control loop is grid connection and stand-alone function. The advantages of this operation modeare higher power reliability and higher power quality.

6. SIMULATION RESULTS

In this section, simulation results under different terms of operation use with Matlab /Simulink ispresented. System block diagram is shown in Figure 10. Detailed model descriptions are given inAppendix A.

35


Figure 6. Shown the output of the neural network by fallowing: (a) The output of the neural network training withthe amount of target data; (b) The output of the neural network of Vmpp with the amount of data; (c) totalerror percentage of the Vmpp; (d) The output of the neural network of MPP with the amount of targetdata;(e) total error percentage of the MPP; (f) Train output versus target data.

6.1 Variation of Irradiance and Temperature

In order to compare the accuracy and efficiency of the three MPPT algorithms selected in this paper,Matlab/Simulink is used to implement the tasks of modeling and simulation. The main objective of thiscase is investigated comparative study of MPPT algorithms under variations of irradiance and temperaturein PV system. The system is connected to the main grid that includes 3200W photovoltaic system and theamount of load is 3200 W. There is no power exchange between photovoltaic system and grid in normalcondition.

The following simulation is presented for different insolation levels at fixed temperature of 25oC asshown in Figure 11(a). The output voltage and the current of PV are depicted in Figure 11 (b) and (c),respectively. When irradiance is decreased at t=3.5 and t=7, it lead to decrease in the output currentof PV as shown in Figure 11(c). It is worth to mention that the evaluation of the proposed controlleris compared and analyzed with the fuzzy logic controller and P&O algorithms. The proposed MPPT

36


Figure 7. shown the output of the neural network test by following: (a) The output of the neural network test withthe amount of target data; (b) The output of the neural network test of Vmpp with the amount of test targetdata; (c) Percentage error of test data Vmpp; (d) The output of the neural network test of MPP with theamount of target data; (e) Percentage error of MPP test data. (f) Test output versus target

Figure 8. FLC structure: (a) Fuzzy controller diagram; (b) Error; (c) Change of Error; (d) Duty cycle.

algorithm can track accurately the MPP when the irradiance changes continuously; also, this method haswell regulated PV output power and it produces extra power rather than fuzzy logic and P&O methodsas indicated in Figure 11(d). Therefore, the injected power from main grid to photovoltaic system isdecreased as demonstrated in Figure 11(e). Fuzzy logic and P&O method perform a fluctuated PV power

37


Figure 9. The inverter control model.

Figure 10. Case study system.

even after the MPP operating has been successfully tracked. Table 4 shows the comparison of real powervalue and presented methods in the different irradiation conditions.

Table 4. Output power values of solar array (watt) in various irradiation conditions

Irradiance Real value ANN+GA Fuzzy logic P&O

0s to 3s 3200 3190 3181 29803s to 7s 2752 2743 2727 26297s to 12s 1132 1124 1109 1005

In order to realize a precise analysis of the performance of the ANN-GA technique, different temperaturelevels at fixed insolation of 1000W/m2 as shown in Figure 12(a). The grid voltage is indicated in Figure12(b). Figure 12(c) shows the variation of the output current of PV. The ANN-GA method shows smotherpower, less oscillating and better stable operating point than fuzzy logic and P&O methods. It has moreaccuracy for operating at MPP also, it generates exceeding power and it possesses faster dynamic responserather than mentioned technique as depicted in Figure 12(d). Consequently, the grid power injectionto the photovoltaic system is declined as illustrated in Figure 12(e). In the view of power stabilization,

38


Figure 11. Simulated results for PV (Variation of Irradiance) in case 1: (a) Irradiance; (b) Inverter output voltage;(c) Inverter output current; (d) PV power; (e) Grid power

the PV power which is controlled by ANN-GA is more stable than the fuzzy logic and P&O methods.Table 5 shows the comparison of real power value and presented methods in the different temperatureconditions.

Table 5. Output power values of solar array (watt) in various temperature conditions

Temperature Real value ANN+GA Fuzzy logic P&O

0s to 4s 3200 3191 3179 29814s to 8s 1235 1226 1212 1108

8s to 12s 2127 2116 2094 1986

From the results, it is noted that the proposed ANN-GA algorithm for MPP shows smother powersignal line and better stable operating point than both fuzzy logic and P&O algorithms. It can be derivedthat the ANN-GA based algorithm has better performance than both fuzzy logic and P&O methods, and ithas more accuracy for operating at MPP.

39


Figure 12. Simulated results for PV (Variation of Temperature) in case 1: (a) Temperature; (b) Grid voltage; (c)Inverter output current; (d) PV power; (e) Grid power.

Figure 13. Simulated results for in case 1: (a) noisy irradiance; (b) PV power; (c) Noisy temperature; (d) PV power.

40


6.2 Noisy Irradiance and Temperature

For realistic conditions, it is needed to analyze the noises in a PV system. Simulation results are shownthe comparison of the proposed method and the fuzzy logic and the P&O methods for irradiation andtemperature changes in different noise levels as illustrated in Figure 13 (a) and (c). It is worth to mentionthat the evaluation of the proposed controller shows the better performance in severe condition thanthe aforementioned methods. At the presence of noise, the tracked output power by using ANN-GA ismore than aforementioned methods, since the proposed method can decrease the effects of uncertainty.Performance of systems by using proposed method is desirable even with limited changes in systemparameters. But using of conventional methods lead to wide error in this situation as depicted in Figure13 (b) and (d).

7. CONCLUSIONS

An integrated scheme for optimal power tracking was proposed in this paper. With the aid of thismethod, the PV system was able to perform and to enhance the production of the electrical energy atan optimal solution under various operating conditions. The GA based offline trained ANN is usedto provide the reference voltage corresponding to the maximum power for any environmental changes.The simulation results show that using ANN-GA controller can dramatically reduce the disadvantagesof previous approaches and also, it can decrease oscillations of power output around the MPP and canincrease convergence speed to achieve the MPP in comparison with fuzzy logic method; also, this methodhad well regulated PV output power and it produced extra power rather than fuzzy logic for differentconditions. The proposed algorithm was verified and it was found that the error percentage of Vmppbetween 0.2% to 0.3%. This error could be reduced by increasing the number of the training data forANN.

Appendix A: Description of the Detailed Model

PV parameters: output power = 3.2kW, Carrier frequency in VMPPT PWM generator: 4.3 kHz and ingrid-Sid controller: 5 kHz, boost converter parameters: L= 3.5mH, C= 630µF, PI coefficients in grid-sidecontroller: KpV dc= 3.5, KiV dc= 7.3, KpId= 8.4, KiId= 343, KpIq= 8.4, KiIq= 343.

References

[1] A. Rezvani, M. Gandomkar, M. Izadbakhsh, and A. Ahmadi, “Environmental/economic schedulingof a micro-grid with renewable energy resources,” Journal of Cleaner Production, 2014.

[2] M. Izadbakhsh, M. Gandomkar, A. Rezvani, and A. Ahmadi, “Short-term resource scheduling of arenewable energy based micro grid,” Renewable Energy, vol. 75, pp. 598–606, 2015.

[3] V. Salas, E. Olias, A. Barrado, and A. Lazaro, “Review of the maximum power point trackingalgorithms for stand-alone photovoltaic systems,” Solar Energy Materials and Solar Cells, vol. 90,no. 11, pp. 1555–1578, 2006.

[4] N. Femia, G. Petrone, G. Spagnuolo, and M. Vitelli, “Optimization of perturb and observe maximum41


power point tracking method,” Power Electronics, IEEE Transactions on, vol. 20, no. 4, pp. 963–973,2005.

[5] C.-C. Chu and C.-L. Chen, “Robust maximum power point tracking method for photovoltaic cells: asliding mode control approach,” Solar Energy, vol. 83, no. 8, pp. 1370–1378, 2009.

[6] F. Liu, S. Duan, F. Liu, B. Liu, and Y. Kang, “A variable step size INC MPPT method for PVsystems,” Industrial Electronics, IEEE Transactions on, vol. 55, no. 7, pp. 2622–2628, 2008.

[7] M. Masoum and M. Sarvi, “An Optimal Fuzzy MPP Tracker Enhancing Battery Charging Perfor-mance of Photovoltaic Systems,”

[8] F. Bouchafaa, I. Hamzaoui, and A. Hadjammar, “Fuzzy Logic Control for the tracking of maximumpower point of a PV system,” Energy Procedia, vol. 6, no. 1, pp. 152–159, 2011.

[9] M. Veerachary, T. Senjyu, and K. Uezato, “Neural-network-based maximum-power-point tracking ofcoupled inductor interleaved-boost-converter-supplied PV system using fuzzy controller,” IndustrialElectronics, IEEE Transactions on, vol. 50, no. 4, pp. 749–758, 2003.

[10] A. K. Rai, N. Kaushika, B. Singh, and N. Agarwal, “Simulation model of ANN based maximumpower point tracking controller for solar PV system,” Solar Energy Materials and Solar Cells, vol. 95,no. 2, pp. 773–778, 2011.

[11] A. Kulaksz, ANN-based control of a PV system with maximum power point tracker and SVM inverter.PhD thesis, 2007.

[12] F. Liu, S. Duan, F. Liu, B. Liu, and Y. Kang, “A Variable Step Size INC MPPT Method for PVSystems,” Industrial Electronics, IEEE Transactions on, vol. 55, no. 7, pp. 2622–2628, 2008.

[13] T. Esram and P. L. Chapman, “Comparison of photovoltaic array maximum power point trackingtechniques,” IEEE Transactions on Energy Conversion, vol. 22, no. 2, pp. 439–449, 2007.

[14] M. G. Simoes and N. Franceschetti, “Fuzzy Optimization Based Control Of A Solar Array System,”in Electric Power Applications, IEE Proceedings-, vol. 146, pp. 552–558, IET, 1999.

[15] D. Huh and M. Ropp, “Comparative Study of Maximum PowerPoint tacking Algorithm Using anExperimental, Programmable, Maximum Power Point Tracking Test Bed,” in Proceedings of 28thIEEE Photovoltaic Specialists Conference, Alaska, pp. 1699–1702, 2000.

[16] T. Hiyama, S. Kouzuma, T. Imakubo, and T. Ortmeyer, “Evaluation of Neural Network Based RealTime Maximum Power Tracking Controller Far PV System,” Energy Conversion, IEEE Transactionson, vol. 10, no. 3, pp. 543–548, 1995.

[17] T. Hiyama and K. Kitabayashi, “Neural Network Based Estimation of Maximum Power Generationfrom PV Module Using Environment Information,” Energy Conversion, IEEE Transactions on,vol. 12, no. 3, pp. 241–247, 1997.

[18] E. Karatepe, M. Boztepe, and M. Colak, “Neural network based solar cell model,” Energy Conversionand Management, vol. 47, no. 9, pp. 1159–1178, 2006.

[19] A. Rezvani, M. Gandomkar, M. Izadbakhsh, and S. Vafaei, “Optimal power tracker for photovoltaicsystem using ann-ga,” International Journal of Current Life Sciences, vol. 4, no. 9, pp. 107–111,2014.

[20] M. R. Vincheh, A. Kargar, and G. A. Markadeh, “A Hybrid Control Method for Maximum PowerPoint Tracking (MPPT) in Photovoltaic Systems,” Arabian Journal for Science and Engineering,vol. 39, no. 6, pp. 4715–4725, 2014.

[21] M. Izadbakhsh, M. Gandomkar, and A. Rezvani, “Improvement of microgrid dynamic performanceunder fault circumstances using ANFIS for fast varying solar radiation and fuzzy logic controller forwind system,” ARCHIVES OF ELECTRICAL ENGINEERING, vol. 63, no. 4, pp. 551–578, 2014.

[22] J. Yang and V. Honavar, “Feature subset selection using a genetic algorithm,” IEEE IntelligentSystems, vol. 13, no. 2, pp. 44–49, 1998.

[23] F. Blaabjerg, R. Teodorescu, M. Liserre, and A. V. Timbus, “Overview of Control and Grid Synchro-42


nization for Distributed Power Generation Systems,” Industrial Electronics, IEEE Transactions on,vol. 53, no. 5, pp. 1398–1409, 2006.

43

SOP 1

Documents

electrical power

solar power

power transmission

extra power

generated power

maximum output power

regulated pv output

electrical energy