International Journal of Scientific & Engineering Research, Volume 7, Issue 4, April-2016 ISSN

Transcription

1 Internatonal Journal of Scentfc & Engneerng Research, Volume 7, Issue 4, Aprl-6 ISSN Optmal Maxmum Power Pont Trackng of PV Systems based Genetc- Hybrd Algorthm F. M. Bendary, Ebtsam. M. saed, Wael A. Mohamed, Z. E. Aff Abstract The maxmum power pont trackng (MPPT) technque n the photovoltac (PV) system s used to acheve maxmum power from the solar PV system. In ths context, three MPPT technques, artfcal neural network (ANN), fuzzy logc control (FLC) and adaptve neurofuzzy nference system (), are mplemented and ther performance s nvestgated n terms of effcency and response. And they are developed n MATLAB/Smulnk envronment. Ths system s developed by combnng the models of establshed solar module and DC-DC boost converter wth the genetc algorthm for the three technques. So ths paper presents a new approach based on the genetc algorthm used to perform a constraned tunng technque for the PID parameters to optmze the power output of solar panel. The dynamc model s used to desgn the controller parameters of the conventonal PID controller. The dynamcs of the DC-DC converter s non lnear. Therefore, t s hard to derve desrable performance. Hence, Genetc algorthm s used to optmze the control parameters of the boost converter. In order to obtan the ftness of an ndvdual, Smulnk model of the boost converter s desgned and the genetc algorthm s programmed to search for the optmal control parameters by the MATLAB bult ngatool. The system s smulated under dfferent clmate condtons and MPPT algorthms. Accordng to the comparsons of the smulaton results, t can be observed that the photovoltac smulaton system can track the maxmum power accurately usng the three MPPT algorthms dscussed. Therefore, the nterest s generated to desgn a more effectve and effcent MPPT to acheve maxmum power transfer to the load. Index Terms Adaptve neuro-fuzzy (), Fuzzy logc (FLC), Genetc algorthm (GA- PID) controller, Maxmum power pont trackng (MPPT) of PV system,neural network (NN). INTRODUCTION ue to growng the global energy demand, the consumpton of electrcal energy s contnuously ncreasng all days, ntellgent systems are progressvely used such as neural backs and need enhancements to be more accurate []. Nowa- D over the world. The energy generaton s mostly dependent on fossl fuels causng the ncrease of greenhouse emsal Network (ANN), Fuzzy Logc Controller (FLC) and adap- network and fuzzy logc MPPT technques [6]. Artfcal Neursons.So; RECENTLY, energy generated from clean, effcent, tve MPPT controllers-based neuro-fuzzy nference system andenvronmentally frendly sources has become one of the () based technques are mplemented [7]. Comparatve majorchallenges forengneers and scentsts. Renewable energy analyss s dscussed among the three dfferent advanced sources are wnd turbne (WT), solar PV system, Bo fuel cell (FC), Bomass and mn hydro power generaton etc. Nowa-days, renewable energy sources are more popular due to varous advantages such as, polluton free, easy avalablty and economc []. Among all renewable energy sources, solar power systems attract more attenton because they provde excellent opportunty to generate electrcty as t s an everlastng, clean renewable energy source and has no potental damage to the envronment. But the effcency of solar cells depends on many factors such as temperature, nsolaton, spectral characterstcs of sunlght, drt, shadow, and so on. Changes n nsolaton on panels due to fast clmatc changes such as cloudy weather and ncrease n ambent temperature can reduce the photovoltac (PV) array output power. However, due to ts low energy converson effcency, many researchers have been focusng to mnmze these drawbacks and ncrease the PV system effcency []. Ths method s commonly named as a maxmum power pont trackng (MPPT) technque. The man functon of MPPT technque s to acheve maxmum power from a PV system [3]. There s a large number of algorthms that are able to track MPP of a PV module have been proposed to solve the problem of effcency. The most common methods are the perturb and observe (P&O) MPPT technques. Genetc Algorthms (GA) are promsng methods for solvng dffcult technologcal problems, and for machne learnng. In ths paper genetc algorthm s used to calculate the optmal control parameters of PID controller of the boost converter [8]. The proposed models show how the GA enhances the model performance when combned wth the prevous three advanced algorthms. The purpose of ths paper s to study and compare advantages, shortcomngs and executon effcency for three power-feedback type MPPT methods, ncludng (ANN), (FLC) and () methods wthout and wth GA-PID controller. Matlab/Smulnk s used n ths paper to mplement the modelng and smulatons tasks, and to compare executon effcency and accuracy for the selected MPPT methods. Ths paper s organzed as follows. Secton s the ntroducton whch ncludes the background of renewable energy and the purpose of ths paper. Secton vews the propsed system confguraton. Sectons 3 and 4 llustrate basc operaton prncples, advantages and shortcomngs for ANN, FLC, Anfs and GA methods respectvely. Secton s the smulaton, analyss and concluson dscusson for the three MPPT algorthms wthout and wth GA-PID. The summary and concluson are gven n Secton 6. and the ncremental conductance (InCond). The frst method s popular due to ts hardware smplcty [4]. The second InCond method has a great accuracy wth good flexblty to rapdly varyng clmatc condtons. But both methods have draw- 6

2 Internatonal Journal of Scentfc & Engneerng Research, Volume 7, Issue 4, Aprl-6 ISSN SYSTEM DESCRIPTION The proposed system s dvded nto four major parts (a) The Solar PV system (b) Power electroncs DC-DC boost converter (c) MPPT technques () Artfcal neural network () Fuzzy logc controller () adaptve neuro-fuzzy nference system. (d) Genetc Algorthm based PID controller. The schematc dagram of the complete system s shown n Fg.. =/(- d) Where I pv s the nput current of the converter, duty cycle d = Ton /T and T=Ton+Toff, wth ts range ( d ). Knowng the V pv and I pv, we can fnd the nput resstance R n of the converter. Ths s gven by (3). R = V /I = R (- (3) n pv pv o d) Here, R n vares from R o to and d vares from to. The locaton of the MPP n the I-V curve of PV module s not known beforehand and always changes dynamcally dependng on rradance and temperature. Therefore, the MPP needs to be located by trackng algorthm, whch s the heart of MPPT controller. The goal of the MPPT s to match the load resstance R L to the optmal output resstance of PV module R opt defned as n (4). R = V /I opt m,pv m, pv (4) When R L matches wth that of Ropt (R n= R opt), the maxmum power transfer from PV to the load wll occur. The MPPT method used n ths work, whch accommodate the duty cycle of the converter, s done by three man technques dscussed n detals n the followng sectons. Fg.. System confguraton dagram.. Photovoltac Cell Modelng An deal solar cell can be modeled as a current source as the solar cell produces current proportonal to the solar rradaton fallng on t. The practcal behavor of a cell s devated from deal due to the optcal and the electrcal losses, so approprate Fg.. Equvalent crcut of photovoltac cell. components should be added wth deal current source. The electrcal crcut representng a solar cell s shown n Fg.. The optcal loss s represented by the current source tself, where thegenerated current I s proportonal to the lght nput. The recombnaton losses are represented by the dode connected parallel to the current source, but n the reverse drecton. The ohmc losses n the cell occur due to the seres and shunt resstance denoted by Rs and Rsh respectvely []. The voltagecurrent characterstc equaton of a solar cell s gven n (). I = I - I exp ((V IR ) /( KT A)) ( V IR ) / R ph s s c s sh () Fg. 3. Boost Converter. Where I ph s a lght-generated current or photocurrent, I s s the cell saturaton of dark current, q= (.6* -9 C) s an electron charge, k= (.38* -3 J/k) s a Boltzmann s constant, Tc s the cell s workng temperature, A s an dealty factor, R sh s a shunt resstance, and R s s a seres resstance. 3 MPPT CONTROL STRATEGIES. Boost Converter Desgn The maxmum power pont trackng s bascally a load matchng problem. In order to change the nput resstance of the panel to match the load resstance, a DC-DC converter s requred (by varyng ts duty cycle). The boost converter s capable of producng a dc output voltage (Vo) greater n magntude than the dc nput voltage (Vs). The crcut topology for a boost converter s as shown n Fg.3. The converson rato s gven n (). V /V = I /I o pv pv o () 6 As mentoned earler, the maxmum power pont of the system vares wth changng condtons. Therefore, the use of MPPT technques s manly essental to extract maxmum power of the PV system. In recent years, ntellgent methods are adopted because of ther reasonng, flexblty and ablty to deal wth the non-lnear and complex system [7]. Hence, n present applcaton artfcal neural network, fuzzy logc and adaptve neurofuzzy nference system based MPPT technques are proposed and dscussed. 3. Neural Network based MPPT Controller Artfcal Neural Network (ANN) s an artfcal network that mmcs the human bologcal neural networks behavor, wdely used n modelng complex relatonshps between nputs and outputs n nonlnear systems. ANN can be defned as pa-

3 Internatonal Journal of Scentfc & Engneerng Research, Volume 7, Issue 4, Aprl-6 ISSN rallel dstrbuted nformaton processng structure consstng of nputs, and at least one hdden layer and one output layer. These layers have processng elements called neurons nterconnected together. An ANN s developed, such that the current solar rradance and temperature are ts nputs and the voltage, whch corresponds to maxmum power, output of the array. The detaled ANN structure and data are as follows;. Collectng Data: The frst step n desgnng an ANN s to collect hstorcal data on the problem that s beng solved usng the network. In case of MPPT lots of array solar rradances and temperatures and ther correspondng maxmum power pont voltages are requred to n order to tran the network.. Selectng Network Structure: The developed ANN n ths thess s two nputs (solar rradance and Temperature) wth two layers (one hdden layer and one output layer). 3. Tranng the Network: The collected tranng ponts are passed nto the desgned network n order to teach t how to perform when dfferent ponts than the tranng ponts are nserted to t usng Matlab. 4. Testng The Network: Some of the test ponts wll be appled n order to fnd out how accurate the developed network s. At the output stage a reference voltage Vmpp s generated to be used n generatng the duty cycle control sgnal for a DC- DC boost converter whch drves the PV voltage to optmal voltage [9]. 3. Fuzzy Logc based MPPT Controller Fuzzy logc controllers have the advantages of workng wth O A x,, (8) mprecse nputs, no need to have accurate mathematcal O B x, 3, 4 model, and t can handle the nonlnearty. It conssts of two nputs and one output. The two FLC nput varables are the Where A and B can adopt any fuzzy membershp error (E), change of error (CE) and output varable s duty functon (MF) []. cycle (D).. Layer : The functon of node s multpled wth every The man parts of a fuzzy logc controller are fuzzfcaton, nput. The output layer declares degree every fuzzy nference, rule base and defuzzfcaton, are shown n Fg.4. E(k) = (PPV (k) - PPV (k -))/(VPV (k) - VPV (k -)) () CE(k) = E(k) - E(k -) (6) Where, P PV s the nstantaneous power of PV array fuzzy nference s processed usng Mamdan s method. Defuzzfcaton uses the center of gravty to process output whch s the duty cycle. n j n D j D j ) / D j D ( (7) j Fg. 4. Block dagram of Fuzzy Controller. The forty nne fuzzy rule base used n ths paper [], s gven n Table based MPPT Controller 3.3. Adaptve neuro-fuzzy prncple: s the hybrd system whch combnes two methods between neural network and FL. Among many FIS models, the Sugeno fuzzy model s the most wdely appled one. For a frst order Sugeno fuzzy model, a common rule set wth two fuzzy f then rules can be expressed as: Rule : f x s A and y s B ; then z = p x + q y +r Rule : f x s A and y s B ; then z = p x + q y +r The conssts of fve layers:. Layer : Every node n the frst layer employ a node functon gven by (8). rule (frng strength ). Equaton n the second layer s as follows; O = = x x, =, A B (9) 3. Layer 3: The -th node n ths layer calculates the rato of the -th rule s frng strength to the sum of all rules frng strengths (normalzed frng strengths ). 3 O = = /( ), =, () 4. Layer 4: In ths layer, every node s adaptve node. Every node s multpled wth p, q, r parameter (Consequentparameters ) []. O 4 = z = (p x + q y + r ) (). Layer :The sngle node n ths layer computes the overall output as the summaton of all ncomng sgnals, whch s expressed as: O TABLE FUZZY RULE BASE E/CE NB NM NS ZE PS PM PB NB ZE ZE ZE NB NB NB NM NM ZE ZE ZE NS NM NM NM NS NS ZE ZE ZE NS NS NS ZE NM NS ZE ZE ZE PS PM PS PS PM PM PS ZE ZE ZE PM PM PM PM ZE ZE ZE ZE PB PB PB PB ZE ZE ZE ZE z z / () 6

4 Internatonal Journal of Scentfc & Engneerng Research, Volume 7, Issue 4, Aprl-6 ISSN The output z n Fg. can be rewrtten as: z = ( x) p + ( y) q + ( y) q + ( ) r Fg.. Archtecture of. + ( ) r + ( x) p (3) 4. Apply some genetc operators (mutaton & crossover) to members of the populaton to create new solutons.. Evaluate these newly created ndvduals. 6. Repeat steps 3,4, and 6 (one generaton) untl the termnaton crtera has been satsfed (usually perform for a certan fxed number of generatons). The concept of mplementaton sequence s the survval of the fttest. The reproductve success of a soluton s drectly ted to the ftness value, whch s assgn durng evaluaton. The least ft soluton may not reproduce at all. The major advantage of GA les n ther computatonal smplcty, and ther powerful search ablty to obtan the global optmum. The further attracton of GA s that they are extremely robust wth respect to complexty of the problem.a smple GA flow chart s shown n Fg Adaptve neuro-fuzzy controller: Desgn of MPPT control usng, to obtan MPP can be acheved wth modfcaton of duty cycle based on changng E(k) and CE(k) n () and (6). FL wth nput and output, algorthm s traned by to track a fuzzy rule. A typcal rule n s gven as follows Rule : f E(k) s U and CE(k) s U then D =y u +y u +y 3 Where, D s duty cycle changng Uj s membershp functon 4 GA BASED PID CONTROLLER 4. Genetc algorthm concept: The genetc algorthm s a method for solvng both constraned and unconstraned optmzaton problems that use 4. Genetc algorthm concept: a performance crteron for evaluaton and a populaton of possble solutons tothe search for a global optmum.the genetc algorthm repeatedly modfes a populaton of ndvdual solutons. At each step, the genetc algorthm selects ndvduals at random from the current populaton to be parents and uses them to produce the chldren for the next generaton. Over successve generatons, the populaton "evolves" toward an optmal soluton. We can apply the genetc algorthm to solve a varety of dffcult optmzaton problems. The genetc algorthm uses three man types of rules at each step to create the next generaton from the current populaton: Selecton. Ths operator selects the ndvduals, called parents, whch contrbute to the populaton at the next generaton. The ftter the ndvdual, the more tmes t s lkely to be selected to reproduce. Crossover.Ths operator randomly combnes two parents to form chldren for the next generaton. Mutaton. Ths operator randomly apples changes to ndvdual parents to form chldren. The process of GA follows ths pattern [8].. Create an ntal populaton (usually randomly generated strng).. Evaluate all of the ndvduals (apply some functon or formula to the ndvduals). 3. Select a new populaton from the old populaton based on the ftness of the ndvduals as gven by the evaluaton functon. Ftness functons used n genetc PID controllers are bascally dependng on the error between the actual and reference predcted solutons. However, better solutons are acheved accordng to decreasng the error sgnal. The goal s to solve some optmzaton problem where we search for an optmal soluton n terms of the varables of the problem (Kp, K and Kd). To mnmze F s equvalent to gettng a mnmum ftness value n the error sgnal. A sutable ftness functon for each set of ndvduals, n ths case, and used n the proposed model expressed by (4). f t 6 Fg. 6. Flow chart of Genetc algorthm. T t e t dt (4) The GA parameters values used n Matlab/smulnk gatool,are as n the followng Table. TABLE GA Property GENETIC ALGORITHM PARAMETERS Populaton sze Maxmum number of generatons Ftness Functon Selecton method Crossover method Value/Method MSE Roulette Crossover propablty 8% Mutaton method Arthmetc Add/Sub Mutaton propablty.%

5 Internatonal Journal of Scentfc & Engneerng Research, Volume 7, Issue 4, Aprl-6 ISSN SIMULATION RESULTS AND DISCUSSION The proposed system was smulated n the MATLAB/Smulnk program. The parameters of the developed photovoltac array are shown n Table 3 Descrpton TABLE 3 PV ARRAY PARAMETERS AT (T= O C AND R=KW/M ) Maxmum Power Voltage at Maxmum Power Current at Maxmum Power Short Crcut Current Open Crcut Voltage Parameter P max= 4.98 W V mp= 7.6 V I mp=.3839 A I SC=. A V OC=.4 V The output power of the solar PV system depends on solar rradaton and temperature. The typcal P-V and V-I characterstcs at dfferent rradaton (R) and temperature (T) are shown n Fgs. 7(a),7b),7(c) and 7(d). Module Current (A) Module Output Power (W) Neural Network based MPPT Controller In desgn procedure of ANN based MPPT controller, solar radaton (R) and ambent temperature (T) are consdered as nputs. A data set s provded for tranng of the network usng NNET tool box n Matlab usng (tranlm) as tranng functon, (tansg) functon transfer for the hdden layer and (pureln) transfer functon for the output layer. Once a neural network s traned, then t can be used to accurately measure optmal voltage for the system at any random set of data whch s not used for tranng. Photovoltac Module P-V Curves R=. NB NM NS ZE PS PM PB R=.4 R=.6.8 R=.8 R= Fg. 7(b). P-V module curves under varous radatons and Tac = o C. Module Current (A) Photovoltac Module I-V Curves Photovoltac Module I-V Curves Tac=c Tac=c Tac=c Tac=7c Tac=c Fg. 7(c). I-V module curves under varous temperatures and R = kw/m. R=. R=.4 R=.6 R=.8 R= Fg. 7(a). I-V module curves under varous radatons and Tac = o C. Module Output Power (W). Fuzzy Logc based MPPT Controller For the fuzzy logc controller desgnng, the nputs are error (E) and change of error (CE) n parameters (voltage and current) of the solar PV system and output s the duty cycle. Ths generated duty cycle (D) targets the DC-DC converter to the optmal voltage. The membershp functons for both nputs and output are shown n Fgs. 8(a), 8(b) and 8(c). Degree of membershp Change n error Fg. 8(a). Input (change n error) Degree of membershp Photovoltac Module P-V Curves NB NM NS ZE PS PM PB -.. Error Fg. 8(b). Input (error) Tac=c Tac=c Tac=c Tac=7c Tac=c Fg. 7(d). P-V module curves under varous temperatures and R = kw/m. 6

6 Internatonal Journal of Scentfc & Engneerng Research, Volume 7, Issue 4, Aprl-6 ISSN \ Degree of membershp.3 based MPPT Controller Usng the Smulnk model of PV module, the operatng solar rradances s vared from W/m to W/m and temperature vared from o C to 74 o C, to get the tranng data sets for. Totally 363 tranng data sets are used to the tranng data and the checkng data. The constructs a FL whose membershp functon parameters are tuned usng hybrd method. The structure of conssts of a fve layers. There are two nputs (temperature and rradance), one output and three membershp functon whch are learned by. In order to study the dynamc response of ANN, FLC and, the desgned MPPT technques, the predefned varyng rradaton and fxed ambent temperature ( C) are consdered as nput to the PV array. As seen n Fg. 9 (a), the solar rradaton vares n the range of W/m to W/m. The comparson of output voltage and power of the three MPPT technques are shown n Fgs.9 (b) and 9(c). The obtaned results show that the performance and effcency of the ANN and are better n comparson of FLC based MPPT technques.also and ANN appeared to be dentcal but there s dfferent whch cleared by calculatons. The trackng effcency of each method s calculated n Table 4. Irradance (W/m) NB NM NS NZ PS PM PB Duty Cycle Fg. 8(c). Output (duty cycle) Tme (Sec) Fg.9(a). The system nput. Fg. 9(b). The PhotoVoltac Output Voltage 3 Fg. 9(c). The PhotoVoltac Output Power PhotoVoltac Output Voltage PhotoVoltac Output Power (Watt) Tme (Sec) FL NN FL NN Tme (Sec).4 based MPPT Controller wthga based PID controller The output voltage and power resulted after GA based PID controller has ncluded n the proposed model, are shown n Fgs.(a), (b). The obtaned results show that the performance and effcency of the s much better after usng GA n determnng the PID varables shown n Table.Also trackng effcency of the combned wth GA based PID controller, s tabulated n Table

7 Internatonal Journal of Scentfc & Engneerng Research, Volume 7, Issue 4, Aprl-6 ISSN TABLE 4 TRACKING EFFICIENCY OF MPPTS Irradaton (W/m ) W/m 7 W/m W/m NN FL + GA P max P avg η pv(%) 98.86% 99.% 99.% P avg η pv(%) 97.% 99.6% 86.7% P avg η pv(%) 99.43% 99.39% 99.34% P avg η pv(%) 99.99% 99.99% 99.99% 6 CONCLUSION In ths paper, an adaptve neuro-fuzzy nference system sproposed for the maxmum power pont trackng under varyng rradance and temperature condtons. The results are obtaned from based MPPT method shows better performances and robustness compared to FL and ANN under varyng rradance condtons. The combned wth PID controller mprovesthe trackng almost to dealty %. GA s very effectve tool that solve dffcult and nonlnear problems n very easy and smple way. PID parameters determned so faster, easer and more accurate than the conventonal methods. PhotoVoltac Output Voltage (Volt) TABLE PID VARIABLES DETERMINED BY GA K P K I K D GA Tme(Sec) Fg. (a) ThePhotoVoltac Output Voltage (+GA) PhotoVoltac Output Power (Watt) GA Tme (Sec) Fg. (b) ThePhotoVoltac Output Power (+GA) REFERENCES [] R. K. Pachaur and Y. K. Chauhan, Hybrd PV/FC Stand Alone Green Power Generaton: A Perspectve for Indan Rural Telecommuncaton Systems, n Proc. IEEE Conference on Issues and Challenges n Intellgent Computng Technques (ICICT), 7-8 Feb. 4 at KIET, Ghazabad, pp [] S. Slvestre, A. Boronat and A. Chouder, Study of Bypass Dodes Confguraton on PV Modules, Appled Energy, vol. 86, no. 9, pp.63-64, Sept. 9. [3] M. A. S. Masoum, H. Dehbone and E. F. Fuchs, "Theoretcal and Expermental Analyses of Photovoltac Systems wth Voltage and Current Based Maxmum Power Pont Trackng," IEEE Trans. OnEnergy Converson., vol. 7, no. 4, pp. 4-, Dec.. [4] B. Smpson, et al, Ttle of paper goes here f known, unpublshed. N. Fema, G. Petrone, G. Spagnuolo, andm. Vtell, Optmzaton of perturb and observe maxmum power pont trackng method, IEEE Trans. Power Electron., vol., no. 4, pp , Jul.. [] Z. Yan, L. Fe, Y. Jnjun, and D. Shanxu, Study on realzng MPPT by mproved ncremental conductance method wth varable step-sze, n Proc. IEEE 3th Ind. Electron. Appl. Conf., 8, pp. 47. [6] Y. C. Kuo, T. J. Lang and J. F. Cben, "Novel Maxmum Power Pont Trackng Controller for Photovoltac Energy Converson System,"IEEE Transactons on Industral Electroncs, vol. 48, no. 3, pp. 94-6, Jun.. [7] C. A. Oteno, G. N. Nyakoe, and C. W. Wekesa, A neural fuzzy based maxmum power pont tracker for a photovoltac system, n Proc. IEEE Conf. (AFRICON 9), Narob, Kenya, Sep. 3, 9, pp. 6. [8] Randy L.Haupt, Sue Ellen Haupt Practcal Genetc Algorthms, Second Edton A JOHN WILEY & SONS, INC., PUBLICA- TION,4. [9] M. Kalamoorthy, R. M. Sekar, I. Raj and G. Chrstopher, Solar Powered Sngle Stage Boost Inverter wth ANN Based MPPT Algorthm, n Proc. IEE conference on Communcaton Control and Computng Technologes (ICCCCT) at Ramanathapuram, 7-9 Oct., pp [] M. A. Islam, A. B. Talukdar, N. Mohammad and P. K. S. Khan, "Maxmum Power Pont Trackng of Photovoltac Arrays n Matlab Usng Fuzzy Logc Controller", Proc. IEEE Inda Conference on Green Energy, Computng and Communcaton (INDICON), 7-9 Dec., at Kolkata, pp. -4. [] A.Tjahjono, O. A. Quds, N. Ayub, W. D. Anggrawan, A. Pryad, M. H. Purnomo Photovoltac Module and Maxmum Power PontTrackng Modellng Usng Adaptve Neuro-FuzzyInference System atmakassar Internatonal Conference on Electrcal Engneerng and Infonnatcs (MICEEI), 6-3 Nov.,