IFOST 2010 Proceedings The New MPPT Algorithm using ANN-Based PV Phan Quoc Dzung HCMC University of Technology Ho Chi Minh City, Vietnam pqdung@hcmut.edu.vn Le DinhKhoa HCMC University of Technology Ho Chi Minh City, Vietnam khoaledinh@hcmut.edu.vn Le Minh Phuong HCMC University of Technology Ho Chi Minh City, Vietnam lmphuong@hcmut.edu.vn Nguyen Truong Dan Vu HCMC University of Technology Ho Chi Minh City, Vietnam ntdnvu@gmail.com Abstract-In grid connected photovoltaic (PV) systems, maximum power point tracking (MPPT) algorithm plays an important role in optimizing the solar energy eiciency In this paper, the new artiicial neural network (ANN) based MPPT method has been proposed for searching maximum power point (PP) fast and This paper presents a new PPT method which decreasing the racking time to reach the PP his is based on combining Artiicial Neural Network for the irst stage nd Incremental Conductnce for he second one exactly For the irst time, the combined method is proposed, II which is established on the ANN-based PV model method and incremental conductance (IncCond) method The advantage of ANN-based PV model method is the fast MPP approximation base on the ability of ANN according the parameters of PV array that used The advantage of IncCond method is the ability to search the exactly MPP based on the feedback voltage and current but don't care the characteristic on PV array The effectiveness of proposed algorithm is validated by simulation using Matlabl Simulink and experimental results using kit ield programmable gate array (FPGA) Virtex II pro ofXilinx Kyword: photovotaic (MPPT), (P), A Incremental Conductance Method [9J The slope of the PV aray power crve is zero at the PP, positive on let of the PP and negative on right of the PP: dP mium power point tracking =0' atMPP dV dP >0' letofMPP (1) dV dP - ' letofMPP (3) V 1V M I < - ' right ofMPP V 1V - - The Incremental Conductnce (lncCond) method is based on hese above equations he PP cn hus be racked by comparing the instantneous conductnce (IN) to the incremental conductnce (1V1V) as shown in the lowchat FigJ Vref is the reference voltage at which the PV aray is forced to operate At the PP, Vref equals to VPP• Once he PP is reached, he operation of the PV rray is maintained at this point nless a change in iradiance, temperature or load In 978-1-4244-9037-0/101$26.00 ©2010 IEEE IFOST 2010 Proceedings those cases, the algorithm decreases or increases Vref to track the new PP here is tradeof in his algorihm Fast PP racking can be achieved wih bigger increments However, he operating point is not remained and oscillates more round PP The New ANN-Model Based MPPT Algorithm As shown above, two hose mehods, ANN and IncCond, have disadvantages while they re used independently he former, ANN mehod, depends on time (for ixed PV aray) and eror of training nd nonlinear unction approximation if the input parmeters different the data that used to train NN The latter, IncCond method, has radeoff between racking time and static eror The combination of two methods solved these problems successully Yes In his method, two stages are used to track PP of PV aray In the irst stage, the trained NNhas guided (Vref,Iref) to optimal point (Vopt,Iopt) which is close to PP he raining data re determined by using Matlab/Simulik simulate the PV aray hought it's pmeters In evey case, which is respect to speciic irradiance nd temperatre values, the crrent-voltage characteristic is recorded for the input data of NN he PP of his chracteristic is also recorded for the output data of ANN The net is implemented for he ppose to determine he voltage of PV aray which the PV rray have he mximum power he net is obtained by raining (supervised) with trainlm unction - Levenberg -Mrquardt algorithm, he acceptable for raining squred eror is 10-2• he optimal number of neurons of 1st layer is 60 logsig neurons, the 2nd layer has purelin neurons So, he total nmber of nerons is 61 neurons (convergence obtained for 1783 epochs) Yes No No In case the prameters of PV rray aren't supported, the IncCond method is used to get he data (V, I, P) that use to train the ANN Figure IncCond Algorithm B Articial Neural Nework Recently, Artiicial Neural Network (ANN) has been strongly developed not only in theory but also in application A common NN showed in Fig2, has many layers: input, hidden nd output layers For PPT, ANN input can be PV aray parmeters like PV voltages nd crrents, environmental data like irradince and temperature, or ny combination of hese The output is usually one or several reference signal(s) like a duty cycle signal used to rive the electronic converter to operate at or close to the PP he input nd output data are rom experimental measurement or model based simulation results Some studies used AN to approximate nonliner unction were published [12, 13] , Input layer Hidden layer Output (11 Proltnptl ytrl {I) � A�itns TrQ: Performace: levenbe