I.

INTRODUCTION

Feedforward Maximum Power Point Tracking of PV Systems Using Fuzzy Controller

MUMMADI VEERACHARY TOMONOBU SENJYU KATSUMI UEZATO University of Ryukyus Japan

A feedforward maximum power (MP) point tracking scheme is developed for the interleaved dual boost (IDB) converter fed photovoltaic (PV) system using fuzzy controller. The tracking algorithm changes the duty ratio of the converter such that the solar cell array (SCA) voltage equals the voltage corresponding to the MP point at that solar insolation. This is done by the feedforward loop, which generates an error signal by comparing the instantaneous array voltage and reference voltage. The reference voltage for the feedforward loop, corresponding to the MP point, is obtained by an off-line trained neural network. Experimental data is used for off-line training of the neural network, which employs back-propagation algorithm. The proposed fuzzy feedforward peak power tracking effectiveness is demonstrated through the simulation and experimental results, and are compared with the conventional proportional plus integral (PI) controller based system. Finally, a comparative study of interleaved boost and conventional boost converter for the PV applications is given and their suitability is discussed.

Manuscript received July 9, 2001; revised February 20, 2002; released for publication June 19, 2002. IEEE Log No. T-AES/38/3/06440. Refereeing of this contribution was handled by I. Batarseh. Authors address: Department of Electrical and Electronics Engineering, Faculty of Engineering, University of the Ryukyus, 1 Senbaru, Nishihara-cho, Nakagami, Okinawa 903-0213, Japan, E-mail: (b985542@tec.u-ryukyu.ac.jp).

c 0018-9251/02/$17.00 2002 IEEE

Photovoltaic (PV) generation is gaining increased importance as a renewable source due to its advantages like absence of fuel cost, little maintenance, no noise and wear due to absence of moving parts, etc. In particular, energy conversion from solar cell array (SCA) received considerable attention in the last two decades. Since the PV generator exhibits a nonlinear v i characteristic, its maximum power (MP) point varies with the solar insolation and temperature. At a particular solar insolation, there is a unique operating point of the PV generator at which its power output is maximum. Therefore, for maximum utilization efficiency, it is necessary to match PV generator to the load such that the equilibrium operating point coincides with the MP point of PV source. However, since the MP point varies with insolation and seasons, it is difficult to maintain MP operation at all solar insolations without changes in the system parameters. To overcome this problem an intermediate dc-dc converter is proposed [13]. The MP point tracking is applied to PV systems in order to extract maximum available power from the SCAs at all solar insolations. Different peak power tracking schemes have been proposed by using different control strategies [49]. Boost converter based MP point tracking using fuzzy logic has been reported [10]. These studies show that the fuzzy control algorithm is capable of improving the tracking performance as compared with the conventional methods. However, all these methods depend on the SCA power output and/or load power detection using the instantaneous voltage and current information, which require voltage and current sensors, respectively. Artificial neural network (ANN) based real time MP tracking controller for PV grid connected systems has been reported [1112]. The studies emphasize that the SCA operating point is shifted to its MP point by using a voltage control type inverter with proportional plus integral (PI) controller, which utilizes the array voltage together with pilot cell voltage information. Array voltage based MP point tracking, using dc-dc converters is yet under investigation. This method of MP tracking has advantages, like straightforward array voltage measurement, no need of current sensors (which introduces losses and complexity in the system), etc. An interleaved dual boost (IDB) converter is proposed here as an intermediate converter (for the PV system) and for tracking the maximum power a fuzzy feedforward controller is developed. Advantages [13] of the present converter system are: 1) ripple cancellation both in the input and output waveforms to maximum extent, 2) lower value of ripple amplitude, high ripple frequency in the resulting input and
JULY 2002 969

IEEE TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS VOL. 38, NO. 3

Authorized licensed use limited to: Jawaharlal Nehru Technological University. Downloaded on October 21, 2009 at 01:28 from IEEE Xplore. Restrictions apply.

output waveforms, and 3) reduced electromagnetic interference because of low ripple amplitude of SCA current. Although the interleaving technique increases the number of components, the actual increase of cost may not be significant. It is because when a large number of converter cells are used they can share the current flow in the inductors and switching devices, so that lower current rating devices can be employed. In the feedforward array voltage based tracking scheme, the MP point tracking depends on the adjustment of reference voltage for the feedforward loop that corresponds to the optimal array voltage at that solar insolation. If the solar insolation changes, then the optimal array voltage also changes. Therefore, an on-line estimation of the optimal array voltage is required for the MP point tracking control. To cope with this situation an off-line ANN is proposed here to estimate the optimal array voltage variation with solar insolation. For controlling the dc-dc converters several control strategies are reported in the literature. These controllers are simple to implement and easy to design. However, there are several drawbacks [14] that hinder the conventional controllers, such as performance dependent on the working point, necessity for tuning of control parameters against changes in supply voltage and load parameters, complex design of control parameters, and stabilization problems, etc. To overcome some of the disadvantages mentioned (above), fuzzy logic controllers (FLCs) are coming up in industrial processes owing to their heuristic nature associated with simplicity and effectiveness for both linear and nonlinear systems. Fuzzy controller application has been successful in many areas, particularly in the field of power electronics to regulate the dc-dc converters [1517], pulsewidth modulated inverters, and lighting of PV powered system [18], etc. This fuzzy control is nonlinear and adaptive in nature, which gives it robust performance under parameter variation, load and supply voltage disturbances, etc. In this work, fuzzy logic has been applied to track the MP from the IDB converter supplied PV system. The control inputs to the FLC are voltage error and change of errors, while the output is the change of control signal for the pulsewidth modulation (PWM) generator. Use of FLCs for the PV systems will relive the burden involved in the design of controller parameters. In addition to this, these controllers will improve the tracking performance as compared with conventional controllers. II. MATHEMATICAL MODEL OF THE SYSTEM The basic configuration of the proposed fuzzy feedforward based MP tracking controller is shown in Fig. 1. Both the controller and the data acquisition system are set up by using PC, A/D and D/A
970

Fig. 1. Experimental setup of PV system.

conversion interfaces, and the other peripherals such as voltage sensing and scaling devices, etc. The analysis of the system is carried out under the following assumptions. 1) Switching elements (MOSFET and Diode) of the converter are assumed to be ideal, i.e., forward voltage drops and ON-state resistances of the switches are neglected. 2) The equivalent series resistance of the capacitance and stray capacitances are neglected. 3) Passive components of the converter (R, L, C) are assumed to be linear, time invariant, and frequency independent. 4) The two converter cells operate in the continuous inductor current mode. 5) The switches (S1 , S2 ) operate in an interleaved fashion. Mathematical models for the individual components are developed in the following sections. A. PV Generator Model The PV generator is formed by the combination of many PV cells connected in series and parallel fashion to provide desired value of output voltage and current. This PV generator exhibits a nonlinear insolation dependent v i characteristic, mathematically expressed for the SCA [2] consisting of Ns cells in series and Np cells in parallel as ! ( ) Ns Ns k IA + (1) V = IA Rs ln 1 + A Np Np I0
JULY 2002

IEEE TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS VOL. 38, NO. 3

Authorized licensed use limited to: Jawaharlal Nehru Technological University. Downloaded on October 21, 2009 at 01:28 from IEEE Xplore. Restrictions apply.

where = (q=AKT), k = Np Iph , qelectric charge; Acompletion factor; KBoltzmanns constant; Tabsolute temperature; Rs cell series resistance; solar insolation; Iph photo current; I0 cell reverse saturation current; IA , V are the SCA current and A voltage, respectively. For given values of SCA parameters, the v i characteristic depends on the solar insolation and the MP point varies with the solar insolation. Rewriting (1) as
V = IA Rs A Ns Np + Ns ln Ipha I0 + ln 1 IA Np Ipha

(2) where Ipha = Iph + I0 . Expanding the term ln(1 IA = Np Ipha ) into Taylor series and neglecting higher order terms results in the following simplified v i characteristic equation ! Ipha Ns 2Ns V = IA Rsg + + ln A Np Ipha I0 (3) where Rsg = Ns Rs =Np . The above equation is used in the simulation studies for realizing the PV source. B. IDB Converter Model The intermediate IDB converter produces a chopped output dc voltage and controls the average dc voltage applied to the load. Further, the converter continuously matches the output characteristic of the PV generator to the input characteristic of the load so that MP is extracted from the SCA. The mathematical model of the IDB converter operating in continuous

current mode is derived using state-space averaging technique. In a more general case for this converter system several topologies are possible depending on the control signals, switching frequency, and load value. However, the analysis is presented for one such case. The theory for the remaining cases should be quite similar to the discussion given here. For the assumed general case, four different modes are possible in one cycle of operation. The corresponding operating modes are, Mode-1: 0 < t d1 T the switches S1 , Dp2 are conducting; Mode-2: d1 T < t (d1 + d2 )T the switches Dp1 , Dp2 are conducting; Mode-3: (d1 + d2 )T < t (d1 + d2 + d3 )T the switches S2 , Dp1 are conducting; Mode-4: (d1 + d2 + d3 )T < t T the switches Dp2 , Dp1 are conducting. The state-space equations for these four different modes of operation, respectively, are (see Appendix for detailed equations) _ X1 = [A1 ][X] + [B1 ][U] _ X2 = [A2 ][X] + [B2 ][U] _ X3 = [A3 ][X] + [B3 ][U] _ X4 = [A4 ][X] + [B4 ][U]: Taking the average of the above four state models results in the following average state-space model: _ X = [A][X] + [B][U] where A = A1 d1 + A2 d2 + A3 d3 + A4 d4 , B = B1 d1 + B2 d2 + B3 d3 + B4 d4 , and d1 + d2 + d3 + d4 = 1. The corresponding A and B matrices are (8) (4) (5) (6) (7)

r1 d2 + d3 + d4 (d + d2 + d3 + d4 ) 0 L1 6 L1 1 6 6 6 r2 d1 + d2 + d4 6 0 (d1 + d2 + d3 + d4 ) 6 L2 L2 6 6 [A] = 6 6 (d + d2 + d3 + d4 ) (d1 + d2 + d4 ) (d2 + d3 + d4 ) 6 1 6 CL CL RCL 6 6 4 (d + d2 + d3 + d4 ) (D + d2 + d3 + d4 ) 1 1 0 CA CA 2 2 0 3 2 i1 3

d1 + d2 + d3 + d4 L1

7 7 7 d1 + d2 + d3 + d4 7 7 7 L1 7 7 7 7 7 0 7 7 7 5 0

6 [B] = 6 4

7 0 7; (d1 + d2 + d3 + d4 ) 5 CA

6i 7 6 27 [X] = 6 7 ; 4 v0 5 V A

[U] = [IA ]:

VEERACHARY ET AL.: FEEDFORWARD MAXIMUM POWER POINT TRACKING OF PV SYSTEMS

971

Authorized licensed use limited to: Jawaharlal Nehru Technological University. Downloaded on October 21, 2009 at 01:28 from IEEE Xplore. Restrictions apply.

Assuming identical branches (r1 = r2 = r, L1 = L2 = L) results in d4 = d2 , d3 = d1 . With these assumptions the simplified matrices A and B are 2 r (D 1) 1 3 1 0 L1 L1 L1 7 6 7 6 6 r2 (D 1) 1 7 6 0 7 6 L2 L2 L2 7 7 6 [A] = 6 7 7 6 (1 D) (1 D) 1 6 0 7 7 6 C CL RCL L 7 6 5 4 1 1 0 0 CA CA 2 3 0 6 7 6 0 7 6 7 [B] = 6 0 7 : 6 7 4 1 5 CA The duty ratio control law is chosen as _ D = f(V , ): A For MP tracking the above control law becomes _ D = K(V V )jVref =Vm : ref A (10) (9)

Fig. 2. Membership functions for E, E, U.

can handle nonlinearity; and 4) they are more robust than conventional nonlinear controllers. This section briefly describes the techniques used in FLC viz. fuzzification, fuzzy knowledge base, and defuzzification. The basic structure of a fuzzy controller used in this work is shown in Fig. 1. In the fuzzification process the numerical variable is converted into a linguistic variable. The following five fuzzy levels are chosen for the controlling inputs of the fuzzy controller (error: E, change of error: E) in the fuzzification: NB: negative big NS: negative small ZE: zero PS: positive small PB: positive big. Membership functions for controller inputs, i.e., E, E, and incremental change in the controller output (U) are defined on the common normalized range of [1, 1]. In this work symmetric triangles with equal base and 50% overlap with other neighboring membership functions are considered. The final membership functions are shown in Fig. 2. A. Fuzzy Knowledge Base The rule base that associates the fuzzy output to the fuzzy inputs is derived by understanding the system behavior. It basically contains the knowledge acquired by the designer as fuzzy rules and is expressed in the form of IF-THEN rules. The fuzzy rules are designed here to incorporate the following considerations keeping in view the overall control performance. 1) When the SCA terminal voltage is much greater than the MP point voltage (V ), then change the duty m ratio (increase) of the converter so as to bring the terminal voltage to V . m 2) When the SCA terminal voltage is less than the MP point voltage, then the change of duty ratio is negative and it must be large so as to bring the terminal voltage to V . m 3) When the array voltage is close to the V , then m incremental duty ratio is small. 4) When the array voltage is near to MP point voltage and is approaching it rapidly, then the change
JULY 2002

Integrating the above question, then the resulting duty ratio in the form of PI control law is Z D = Kp (V V ) + Ki (V V )dt: (11) ref A ref A In real time the duty ratio for the converter is generated by comparing a triangular carrier signal (V ) tr with control (or) reference signal (V ), i.e., the duty c ratio of the converter is V (12) D= c: V tr For a given triangular carrier signal the duty ratio is proportional to the control signal (V ). Therefore, c the control law (12) becomes Z V = Kp (V V ) + Ki (V V )dt: (13) c ref A ref A The above control law is implemented in real-time using the following discrete form X (V (k) V (k)): V (k) = Kp (V (k) V (k)) + Ki c ref A ref A (14) III. FUZZY CONTROLLER In recent years, FLCs [1419] have been widely used for industrial processes owing to their heuristic nature associated with simplicity and effectiveness for both linear and nonlinear systems. Advantages of FLCs over the conventional controllers are: 1) they do not need accurate mathematical model; 2) they can work with imprecise inputs; 3) they
972

IEEE TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS VOL. 38, NO. 3

Authorized licensed use limited to: Jawaharlal Nehru Technological University. Downloaded on October 21, 2009 at 01:28 from IEEE Xplore. Restrictions apply.

TABLE I Rule Base Table for FLC

to bring the terminal voltage to the desired value. In the following sections the method of estimating the reference voltage for the feedforward loop using neural networks is discussed. IV. ARTIFICIAL NEURAL NETWORK ANNs are widely accepted as a technology offering an alternative way to solve complex problems. Particularly, in recent years the application of ANN models in various fields is increasing because, these ANNs operate like a black box model, requiring no detailed information about the system. They learn the relationship between the input and output variables by studying the previously recorded data. Another advantage of using ANNs is their ability to handle large and complex systems with many interrelated parameters. These trained ANNs can be used to approximate an arbitrary input-output mapping of the system. In the voltage based peak power tracking scheme, the reference voltage to the feedforward loop is to be adjusted such that it is equal to the MP point voltage at that solar insolation. Since the solar insolation is varying, the corresponding reference voltage (V = V ) for the feedforward loop should ref m also change according to the insolation variation. Therefore, for MP point tracking control, an on-line estimation of the reference voltage for the feedforward loop is essential. Since the MP point voltages are nonlinearly related to the solar insolation, the linear function approximation techniques were not suitable. Further, it may not be possible to find a closed form relationship between V and . Even if we are able m to find the relationship between these two variables for a limited range, employing curve fit methods, they may not result in true MP point voltages for the whole range of operation. Thus, the conventional curve fit methods were not suitable to realize the true peak power trackers. Under these circumstances, the ANNs provide a viable solution for the on-line estimation of the insolation-dependent reference voltage. In these studies a three layer feedforward neural network with sigmoid activation function is considered for the on-line estimation of reference voltage. Among the available training algorithms, the back-propagation algorithm [20] is one of the most widely used, because it is stable, robust, and easy to implement. We now begin by considering the feedforward neural network consisting of a single hidden layer with sigmoid activation function. An input vector, x = = (1 , 2 , 3 : : : , n )T is applied to the input layer of the network as shown in Fig. 3. The net input of the jth hidden unit is neth = j
n X i=1 h wji xi + j

of duty ratio should be zero so as to prevent operating point deviation away from the MP point. 5) When the array voltage is equal to the MP point voltage, then the change of duty ratio should be maintained at zero. Taking the above points into consideration the fuzzy rules are derived and the corresponding rule base is given in Table I. There are several possible combinations of the degree of supports with varying strengths to the corresponding rules, to satisfy different conditions. However, for the present problem one such combination for the degree of support, resulting in better tracking performance is 1.0, 0.5, 0.3, 0.0. B. Defuzzification In the defuzzification process the crisp value of the change of duty cycle is obtained. The well-known center of gravity [19] method for defuzzification is used here. It computes the center of gravity from the final fuzzy space, and yields a result which is highly related to all of the elements in the same fuzzy set. The crisp value of control output U is computed by the following equation Pn i i=1 W U U = Pn i (15) i i=1 W where n is the maximum number of effective rules, W is the weighting factor, and Ui is the value i corresponding to the membership function of U. Using the steps mentioned above, the fuzzy controller is implemented in real-time for MP point tracking. The data acquisition system samples the terminal and reference voltages of the feedforward loop and computes error E = (V V ), change of ref A error signals at each sampling time. Employing these error and change of error signals, the fuzzy controller determines the control action required from the fuzzy knowledge base. Then it computes the required change in the control voltage for the PWM generator, which changes the duty ratio of the converter so as

(16)

VEERACHARY ET AL.: FEEDFORWARD MAXIMUM POWER POINT TRACKING OF PV SYSTEMS

973

Authorized licensed use limited to: Jawaharlal Nehru Technological University. Downloaded on October 21, 2009 at 01:28 from IEEE Xplore. Restrictions apply.

ANN training is performed. The steps involved in the training process are as follows. Step 1 Initialize the network synaptic weights with small random values. Step 2 Apply an input vector (solar insolation) to the network and calculate the corresponding output values. Step 3 Compare the actual outputs with the desired outputs (reference voltage) and determine the measure of error (evaluating function). Step 4 Determine the amount by which each weight is to be changed (16)(22) and update all the connection weights. Step 5 Repeat steps 2 to 4 with all training vectors until the error for the vectors in the training set is reduced to an acceptable value. V. EXPERIMENTAL SYSTEM DESCRIPTION

Fig. 3. Feedforward neural network.

where wji is the weight on the connection from the h ith input unit, j for j = 1, 2, : : : represents the bias for hidden layer neurons, which is mainly used to improve the learning speed during network training process. Now, the output of the neurons in the hidden layer is written as n ! X h h yj = f wji xi + j (17)
i=1

and the net input to the neurons in the output layer becomes Nh X h 0 wkj yj + k (18) net0 = k
j=1 h where j represents the bias for neurons in the output layer. Finally, the output of the neurons (reference voltage for the feedforward loopV ) in the output ref layer is 0 1 Nh X 0 h 0 yk = f @ wkj yj + k A : j=1

(19)

The learning stage of the network is performed by updating the weights and biases using back-propagation algorithm with the gradient descent method in order to minimize a mean squared error performance index Ep given as Ep = 1 (V (n) V (n))2 : A 2 (ref) The synaptic weights updating expressions are ! @Ep + wji (n) wji (n + 1) = wji (n) @wji (n) (21) wji (n) = wji (n) wji (n 1) (22) (20)

The basic configuration of the proposed PV system is shown in Fig. 1. The data acquisition system is setup by using PC, interface AZI-3503 card, which mainly consists of 8-channel 12-bit A/D, D/A converters. For power measurements a digital power meter (YOKOGAWA-WT130) is used, through which a GPIB interface is connected to the PC to record the SCA power data. The PWM modulator is a voltage comparator made of LF311 operational amplifier. The reference signal to this comparator is the signal obtained from the D/A converter, generated by means of the MPPT algorithm. A synthesized YOKOGAWA function generator (FG120) was used to obtain phase displaced triangular carrier signals to the PWM generator. The experimental prototype circuit was built with an International Rectifier IRF530 MOSFET with suitable driver circuit, and the diode FML-32S. The artificial sun is realized in the laboratory by means of incandescent lamp set. Further, the solar insolation level illuminated on the solar panel is adjusted by controlling the power to this incandescent lamp set. VI. EXPERIMENTAL RESULTS AND DISCUSSIONS Comprehensive simulation studies were made to investigate the feedforward voltage based MP point tracking capability of IDB converter supplied PV system. To verify the theoretical analysis and modelling equations developed in the previous sections, a design example was considered. The SCA and converter parameters are given in Tables II and III, respectively. The PV array is simulated using the equation (3), whereas the converter and duty ratio control are defined by (4)(8) and (14). These sets of equations are programmed in the continuous dynamic system simulator (TUTSIM). The simulated peak power tracking characteristics for one solar insolation 1 are shown in Figs. 6 to 8. To validate
JULY 2002

where , are learning and momentum factors, respectively. The network training is performed repeatedly until the performance index Ep = (V V )2 ref A reduces below a specified value, ideally to zero. In other words when Ep ! 0 leads to (V V )2 ! 0, ref A then the trained neural network connecting weights are adjusted in such a way that the estimated array voltage is identically equal to the MP point voltage. With the above equations in the forward direction, the
974

IEEE TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS VOL. 38, NO. 3

Authorized licensed use limited to: Jawaharlal Nehru Technological University. Downloaded on October 21, 2009 at 01:28 from IEEE Xplore. Restrictions apply.

TABLE II PV Array Parameters Maximum power (Pm ) Open circuit voltage (V ) oc Short circuit current (Isc ) Operating voltage at maximum power (V ) m Operating current at maximum power (Im ) Module efficiency Module size 24 W 20.7 V 2.2 A 12.35 V 1.94 A 11% 9:85 4:45 cm

(AM 1.5, 1 kW/m2 , 25 C) TABLE III Converter Parameters r1 r2 L1 L2 Ca C 0:064 0:073 250 H 250 H 47 F 10 F

Fig. 4. Experimental V IA =V Pg characteristics of SCA A A module for different solar insolations.

the simulation results, experimental investigations are presented in the following paragraphs. A prototype PV supplied converter system was constructed to study the array voltage based MP point tracking using fuzzy controller. The experimental V IA , V Pg characteristics of the PV generator, A A obtained at different solar insolations are shown in Fig. 4. Fo real-time MP point tracking, the program was developed in the C environment employing the equations derived in the preceding sections. Solar insolation was measured with the help of insolation meter, which gives the equivalent voltage signal, and the SCA input voltage was measured by a sensing and scaling device. These two voltage signals were sensed by the A/D converter of the data acquisition system. The off-line ANN sets the reference voltage to the feedforward loop from the known solar insolation equivalent voltage signal. At a given solar insolation for MP operation of SCA, the array operating voltage must be made equal to theoptimum voltage V by m changing the converter duty ratio. The duty ratio is controlled by means of a PWM generator control signal (V ), obtained from the fuzzy controller. This c fuzzy controller works based on the controlling inputs, error (E) and change of error (E) signals, generated with the help of feedforward loop. The error signal generated by feedforward loop depends on the instantaneous array voltage (V ) and the reference A voltage (V ). Estimating this reference voltage by ref means of an off-line ANN is given in the following lines. Experiments were conducted on the chosen PV generator, and the MP point voltages (V ) were m determined. Their variation with solar insolation is shown in Fig. 5. Taking these values as reference patterns, the feedforward ANN is trained. Gradient

Fig. 5. Feedforward loop reference voltage variation with solar insolation.

descent algorithm is used in training, as it improves the performance of the ANN, reducing the total error by changing the weights along its gradient. The learning rate parameter is 0.5 and momentum factor of 0.9 is used for satisfactory performance. The accuracy of this trained ANN is verified by considering test points (solar insolations), which include both trained and untrained values. The test results are also plotted in Fig. 5 and they are in close agreement with the experimental results. Now this trained ANN is combined with the peak power tracking scheme for on-line estimation of the reference voltage to the feedforward loop at different solar insolations. The reference voltage variation with solar insolation, estimated from the off-line trained ANN, is also shown in Fig. 5. The developed scheme tracks the MP point continuously by adjusting the SCA terminal voltage to MP point voltage. At 100% solar insolation (1 ) the experimental tracking characteristics are shown in Figs. 68. Studies are also made to observe the effectiveness of the developed tracking scheme for changing solar insolations. Experimental observations are given in Fig. 9 for three discrete levels (2 , 4 , 6 ), other than the ones on which the ANN was trained. These results show that, the
975

VEERACHARY ET AL.: FEEDFORWARD MAXIMUM POWER POINT TRACKING OF PV SYSTEMS

Authorized licensed use limited to: Jawaharlal Nehru Technological University. Downloaded on October 21, 2009 at 01:28 from IEEE Xplore. Restrictions apply.

Fig. 6. Experimental SCA power tracking characteristics at solar insolation 1 .

Fig. 8. Error in terminal voltage tracking at solar insolation 1 .

Fig. 7. Experimental SCA terminal voltage tracking characteristics at solar insolation 1 .

Fig. 9. Experimental SCA power tracking characteristics for discrete solar insolations.

developed algorithm is capable of tracking MP point even for variable solar insolations. The experimental array power tracking characteristics for different solar insolations (1 , 3 , 5 , and 7 ) are also obtained as shown in Fig. 10. For verification of MP points of SCA, experiments were conducted on the SCA by connecting a variable load resistance. The experimental MP points obtained at different solar insolations are tabulated in Table IV. Comparing the tracking characteristics (Fig. 10) with the MP points, it can be noticed that, the feedforward control strategy is capable of extracting MP from the SCA. The tracking capability of the IDB converter system is also verified under partial shading conditions. For illustration, array power tracking characteristics, when four cells shaded by 50% are shown in Fig. 11. Under this condition, the SCA power output decreases and settles to a new MP point as shown in Fig. 11. The proposed fuzzy controller based scheme is evaluated by comparing its tracking performance with that of a conventional PI controller based scheme. Experimental studies are made on the IDB converter supplied PV system with conventional PI controller. For comparison purpose, the experimental MP tracking characteristics with fuzzy and conventional
976

Fig. 10. Experimental SCA power tracking characteristics for different shading condition.

PI controller are shown in Figs. 610. It can be noticed from these characteristics that, fuzzy controller improves the tracking performance over the conventional PI controller based PV system. In particular, for variable solar insolations the fuzzy controller provides better performance compared with PI controller. Under variable solar insolation conditions, if conventional PI controller is used for achieving the optimum tracking performance, it requires tuning of the controller
JULY 2002

IEEE TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS VOL. 38, NO. 3

Authorized licensed use limited to: Jawaharlal Nehru Technological University. Downloaded on October 21, 2009 at 01:28 from IEEE Xplore. Restrictions apply.

Fig. 11. Experimental SCA power tracking characteristics for partial shading condition.

Fig. 12. Experimental SCA power tracking characteristics for battery load.

TABLE IV Experimental MP Points Solar Insolation 1 2 3 4 5 6 7 8 Maximum Power (W) 23.03 19.07 17.04 15.55 13.11 11.03 8.05 4.63 Fig. 13. Experimental tracking characteristics of the SCA for dynamic load changes.

gains appropriately depending on the solar insolation. From the experimental results, it is observed that the feedforward control strategy with fuzzy controller is a promising one with reference to peak power tracking. Furthermore, it does not require any tuning of the parameters, which is the case with conventional PI controller, wherein the controller gain parameters needs to be changed when solar insolation changes. The validity of the proposed tracking scheme is also verified for battery charging application. For illustration, experimental results are presented for one particular solar insolation (1 ) in Fig. 12. From these observations it is understood that, in this case also the fuzzy control improves the power tracking performance as compared with PI controller based power tracking. The tracking effectiveness is also verified for dynamic changes in the load. For illustration, experimental observations are presented for the load change sequence R E, E, R E (R: resistance, E: battery load) in Fig. 13 for two different solar insolations. These results indicate that in the proposed tracking scheme, power extraction from the SCA does not depend on the nature of load connected. It is immaterial whatever may be the load that is connected to the converter, the power extraction (algorithm) depends only on the reference array voltage that is set to the feedforward loop. When the load changes from R E, to E and vice versa, the

converter will only change the duty ratio such that load input characteristic matches with the SCA output characteristic as evidenced by Fig. 13. The above experimental results agree well with those obtained from simulations. However, slight differences in these results (Figs. 68) are attributed to the following factors. 1) The analysis was made on the assumption that all the switching devices are ideal, i.e., in the modeling, nonidealities of the converter such as forward voltage drops, on-state resistances of the switching devices and other parasitics are not taken into account. 2) There are errors in the measuring system, drops in the parasitics, etc. A comparative study of boost and interleaved boost converter supplied PV system operating at MP point tracking conditions was made and their steady-state performance parameters (ripple amplitudes of array current, load voltage, and efficiencies) were measured for different solar insolations. These results are tabulated in Table V for comparison purposes. Tabulated measurements show that the interleaved operation of the parallel boost cells will have substantial effect in reducing the ripple content both on the source and load side, switch current stresses, improving the efficiency of the converter system. The observed reduction in
977

VEERACHARY ET AL.: FEEDFORWARD MAXIMUM POWER POINT TRACKING OF PV SYSTEMS

Authorized licensed use limited to: Jawaharlal Nehru Technological University. Downloaded on October 21, 2009 at 01:28 from IEEE Xplore. Restrictions apply.

Fig. 14. Experimental waveforms of boost converter at MP tracking conditions.

Fig. 16. Comparison of ripple reduction in boost and interleaved boost converters (Ca = 47 F).

Fig. 15. Experimental waveforms of interleaved boost converter at MP tracking conditions.

TABLE V Comparison of Experimental Ripple Amplitudes and Efficiency at MP Point Tracking Conditions IA Solar Insolation 1 3 5 7 8 Boost 1.10 0.96 0.74 0.56 0.33 IDB 0.54 0.48 0.32 0.32 0.23 V 0 Boost 1.92 1.20 1.12 0.88 0.64 IDB 1.52 1.12 0.96 0.64 0.40 Boost 72.08 80.73 84.29 85.94 89.32 IDB 77.53 86.59 88.27 89.49 91.79

Normally, the PV systems require a capacitance to filter the terminal quantities (voltage and current), and its value depends on the magnitude of the ripple to be canceled. In case of interleaved boost converter, the interleaved operation itself reduces the ripple magnitude and hence the terminal capacitance requirement is less as compared with boost converter supplied PV system. To demonstrate this capacitance effect on the array current and voltage ripple reduction, a terminal capacitance of Ca = 47 F is connected in either case. From these observations it can be noticed that, for a given array capacitance, the ripple reduction is more in the interleaved converter over the conventional boost converter supplied PV system. The corresponding experimental waveforms (array voltage and current) are shown in Fig. 16. In addition to the salient points mentioned above, it is observed that the current through the switching devices is comparatively smaller in the interleaved boost converter, i.e., lower switch current stress. This point clearly indicates that lower current rating switching devices are sufficient for the interleaved converter system and thus the increase in the cost of this converter system, due to increase in the number of components, is not significant as compared with conventional boost converter. In view of the advantages discussed, the interleaved converter is preferable for the PV applications. VII. CONCLUSIONS A fuzzy feedforward voltage based MP point tracking scheme has been developed for the IDB converter supplied PV system. Analytical expressions for the PV source and converters are derived. An off-line ANN, trained using back-propagation algorithm, is utilized for on-line estimation of reference voltage for the feedforward loop. Its estimation accuracy is verified with experimental results. Peak power tracking capability of the proposed scheme was demonstrated through simulation and experimental results. Further, it is
JULY 2002

load voltage ripple is about 1535%, while in the array current is 3050%. Furthermore, the interleaved operation also results in improved efficiency. Under peak power tracking conditions the interleaved boost converter exhibits higher efficiency, about 25%, over the conventional boost converter. At one particular solar insolation (1 ) detailed experimental waveforms of PV array current, voltage, and load voltage under MP point tracking conditions are shown in Figs. 14 and 15 for conventional boost and interleaved boost converters, respectively. It is evident from these experimental waveforms, that the interleaved operation results in an improved performance.
978

IEEE TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS VOL. 38, NO. 3

Authorized licensed use limited to: Jawaharlal Nehru Technological University. Downloaded on October 21, 2009 at 01:28 from IEEE Xplore. Restrictions apply.

also demonstrated that the fuzzy control improves the tracking performance compared with the conventional PI controller and thus avoids the tuning of controller parameters. A comparative study with boost converter was made and this study reveals that interleaved converter is preferable for PV applications on account of its salient features discussed in this work. Furthermore, the use of IDB converter avoids the discontinuous input current mode of operation and reduces the ripple in the array input current, terminal capacitance requirement, etc. As a consequence, an improved SCA performance can be achieved. APPENDIX Mode-1: _ {1 = _ {2 = _ v0 = V ri A 11 L1 L1 V ri v A 22 0 L2 L2 L2 i2 v 0 CL RCL (23) (24)
[3] [2]

Mode-4: _ {1 = _ {2 = _ v0 = V ri v A 11 0 L1 L1 L1 V ri v A 22 0 L2 L2 L2 i1 i v + 2 + 0 CL CL RCL (38) (39) (40) (41) (42)

i i I _ V = 1 2 + A A CA CA CA _ X4 = [A4 ][X] + [B4 ][U]:

(25) (26)
[4]

i i I _ V = 1 2 + A A CA CA CA _ X1 = [A1 ][X] + [B1 ][U]: Mode-2: _ {1 = _ {2 = _ v0 = V ri v A 11 0 L1 L1 L1 V ri v A 22 0 L2 L2 L2 i1 i v + 2 0 CL CL RCL

(27)
[5]

(28) (29)

[6]

[7]

(30) (31) (32)

i i I _ V = 1 2 + A A CA CA CA _ X1 = [A2 ][X] + [B2 ][U]: Mode-3: V ri v _ {1 = A 1 1 0 L1 L1 L1 _ {2 = _ v0 = V ri A 22 L2 L2 i1 v 0 CL RCL

[8]

(33) (34) (35)

i i I _ V = 1 2 + A A CA CA CA _ X3 = [A3 ][X] + [B3 ][U]:

(36) (37)

Salameh, Z., and Taylor, D. (1990) Step-up maximum power point tracker for photovoltaic arrays. Solar Energy, 44 (1990), 5761. Alghuwainem, S. M. (1992) Steady-state performance of dc motors supplied from photovoltaic generators with step-up converter. IEEE Transactions on Energy Conversion, 7 (1992), 267272. Appelbaum, J., Singer, S., and Sarma, S. M. (1992) Starting characteristics of permanent magnet and series excited motors powered by solar cells: Variation with solar radiation and temperature. Electric Machines and Power Systems, 20 (1992), 173181. Siri, K., Caliskan, V. A., and Lee, C. Q. (1993) Peak power tracking in parallel connected converters. IEE Proceedings, Pt. G, 140, 2 (1993), 106116. Kislovski, A. S. (1993) Power tracking methods in photovoltaic applications. Proceedings of Power Conversion (June 1993), 513528. Teulings, W. J. A., Marpinard, J. C., and Capel, A. (1993) A maximum power point tracker for a regulated power bus. In Proceedings of the European Space Power Conference, Austria, 1993, 9397. Hua, C., Lin, J., and Shen, C. (1998) Implementation of a DSP controlled photovoltaic system with peak power tracking. IEEE Transactions on Industrial Electronics, 45 (1998), 99107. Matsui, M., Kitano, T., Xu, D-H., and Yang, Z-Q. (2000) New MPPT control scheme utilizing power balance at dc link instead of array power detection. In Proceedings of International Power Electronics Conference (IPEC), 2000, 158163. Sharif, M. F., Alonso, C., and Martinez, A. (2000) A simple and robust maximum power point control for ground photovoltaic generators. In Proceedings of International Power Electronics Conference (IPEC), 2000, 164169. Won, C-Y., Kim, D-H., Kim, S-C., Kim, W-S., and Kim, H-S. (1994) A new maximum power point tracker of photovoltaic arrays using fuzzy controller. In Proceedings of PESC (1994), 396403. Hiyama, T., Kouzuma, S., Imakubo, T., and Ortmeyer, T. H. (1995) Evaluation of neural network based real time maximum power tracking controller for PV system. IEEE Transactions on Energy Conversion, 10 (1995), 543548. 979

VEERACHARY ET AL.: FEEDFORWARD MAXIMUM POWER POINT TRACKING OF PV SYSTEMS

Authorized licensed use limited to: Jawaharlal Nehru Technological University. Downloaded on October 21, 2009 at 01:28 from IEEE Xplore. Restrictions apply.

[12]

[13]

[14]

[15]

[16]

de Medeiros Torres, A., Antunes, F. L. M., and Soares dos Reis, F. (1998) An artificial neural network-based real time maximum power tracking controller for connecting a PV system to the grid. Proceedings of IEEE IECON, 1 (1998), 554558. Giral, R., Martinez-Salamero, L., and Singer, S. (1999) Interleaved converters operation based on CMC. IEEE Transactions on Power Electronics, 14, 4 (1999), 643652. So, W. C., Tse, C. K., and Lee, Y. S. (1994) A fuzzy controller for dc-dc converters. Proceedings of IEEE, PESC (1994), 315320. Gupta, T., Boudreaux, R. R., Nelms, R. M., and Hung, J. Y. (1997) Implementation of a fuzzy controller for dc-dc converters using an inexpensive 8-b microcontroller. IEEE Transactions on Industrial Electronics, 44, 5 (1997), 661669. Mattavelli, P., Rossetto, L., Spiazzi, G., and Tenti, P. (1997) General purpose fuzzy controller for dc-dc converters. IEEE Transactions on Power Electronics, 12, 1 (1997), 7985.

[17]

[18]

[19]

[20]

Raviraj, V. S. C., and Sen, P. C. (1997) Comparative study of proportional-integral, sliding mode, and fuzzy logic controllers for power converters. IEEE Transactions on Industry Applications, 33, 2 (1997), 518524. Wu, T-F., Chang, C-H., and Chen, Y-K. (2000) A fuzzy logic controlled single stage converter for PV powered lighting system applications. IEEE Transactions on Industrial Electronics, 47, 2 (2000), 287296. Harris, C. J., Moore, C. G., and Brown, M. (1993) Intelligent Control: Aspects of Fuzzy Login and Neural Nets. London: World Scientific, 1993. Widrow, B., and Lehar, M. A. (1990) 30 years of adaptive neural networks: Perceptran, madaline, and backpropagation. Proceedings of IEEE, 78 (1990), 14151442.

Mummadi Veerachary was born in Survail, AP, India in 1968. He obtained his Bachelors degree from College of Engineering Anantapur, JNT University, Hyderabad, India, in 1992 and Master of Technology from Regional Engineering College, Warangal, India in 1994. In 1994 he joined, as an Assistant Professor, the Dept. of Electrical Engineering, JNTU College of Engineering, Anantapur, India. Presently, he is at the Dept. of Electrical and Electronics Engineering, University of the Ryukyus, Okinawa, Japan for his research studies. His fields of interest are power electronics, modeling and simulation of power electronics and application to photovoltaic solar energy utilization. Mr. Veerachary was the recipient of the IEEE Industrial Electronics Society student travel grant award for the year 2001.
980 IEEE TRANSACTIONS ON AEROSPACE AND ELECTRONIC SYSTEMS VOL. 38, NO. 3 JULY 2002

Authorized licensed use limited to: Jawaharlal Nehru Technological University. Downloaded on October 21, 2009 at 01:28 from IEEE Xplore. Restrictions apply.

Tomonobu Senjyu was born in Saga prefecture, Japan, in 1963. He received the B.S. and M.S. degrees in electrical engineering from University of the Ryukyus, Okinawa, Japan, in 1986 and 1988, respectively, and the Ph.D. degree in electrical engineering from Nagoya University, Nagoya, Japan, in 1994. Since 1988, he has been with the Department of Electrical and Electronics Engineering, Faculty of Engineering, University of the Ryukyus, where he is currently a Professor. His research interests are in the areas of stability of ac machines, advanced control of electrical machines, and power electronics. Dr. Senjyu is a member of the Institute of Electrical Engineers of Japan.

Katsumi Uezato was born in Okinawa prefecture, Japan, in 1940. He received the B.S. degree in electrical engineering from the University of the Ryukyus, Okinawa, Japan, in 1963, the M.S. degree in electrical engineering from Kagoshima University, Kagoshima, Japan, in 1972, and the Ph.D. degree in electrical engineering from Nagoya University, Nagoya, Japan, in 1983. Since 1972, he has been with the Department of Electrical and Electronics Engineering, Faculty of Engineering, University of the Ryukyus, where he is currently a Professor. He is engaged in research on stability, control of synchronous machines and power electronics. Dr. Uezato is a member of the Institute of Electrical Engineers of Japan.
VEERACHARY ET AL.: FEEDFORWARD MAXIMUM POWER POINT TRACKING OF PV SYSTEMS 981

Authorized licensed use limited to: Jawaharlal Nehru Technological University. Downloaded on October 21, 2009 at 01:28 from IEEE Xplore. Restrictions apply.