A Novel Sensorless Support Vector Regression Based Multi-Stage Algorithm to Track the Maximum Power Point for Photovoltaic Systems

Transcription

1 A Novel Sensorless Support Vector Regresson Based Mult-Stage Algorthm to Track the Maxmum Power Pont for Photovoltac Systems by Ahmad Osman Khedr Ibrahm A thess presented to the Unversty of Waterloo n fulfllment of the thess requrement for the degree of Master of Appled Scence n Electrcal and Computer Engneerng Waterloo, Ontaro, Canada, 2012 Ahmad Osman Khedr Ibrahm 2012

2 AUTHOR'S DECLARATION I hereby declare that I am the sole author of ths thess. Ths s a true copy of the thess, ncludng any requred fnal revsons, as accepted by my examners. I understand that my thess may be made electroncally avalable to the publc.

3 Abstract Solar energy s the energy derved from the sun through the form of solar radaton. Solar powered electrcal generaton reles on photovoltac (PV) systems and heat engnes. These two technologes are wdely used today to provde power to ether standalone loads or for connecton to the power system grd. Maxmum power pont trackng (MPPT) s an essental part of a PV system. Ths s needed n order to extract maxmum power output from a PV array under varyng atmospherc condtons to maxmze the return on ntal nvestments. As such, many MPPT methods have been developed and mplemented ncludng perturb and observe (P&O), ncremental conductance (IC) and Neural Network (NN) based algorthms. Judgng between these technques s based on ther speed of locatng the maxmum power pont (MPP) of a PV array under gven atmospherc condtons, besdes the cost and complexty of mplementng them. The P&O and IC algorthms have a low mplementaton complexty but ther trackng speed s sluggsh. NN based technques are faster than P&O and IC. However, they may not provde the global optmal pont snce they are prone to multple local mnma. To overcome the demerts of the aforementoned methods, support vector regresson (SVR) based strateges have been proposed for the estmaton of solar rradaton (for MPPT). A sgnfcant advantage of SVR based strateges s that t can provde the global optmal pont, unlke NN based methods. In the publshed lterature of SVR based MPPT algorthms, however, researchers have assumed a constant temperature. The assumpton s not plausble n practce as the temperature can vary sgnfcantly durng the day. The temperature varaton, n turn, can remarkably affect the effectveness of the MPPT process; the ncluson of temperature measurements n the process wll add to the cost and complexty of the overall PV system, and t wll also reduce the relablty of the system.

4 The man goal of ths thess s to present a novel sensorless SVR based mult-stage algorthm (MSA) for MPPT n PV systems. The proposed algorthm avods outdoor rradaton and temperature sensors. The proposed MSA conssts of three stages: The frst stage estmates the ntal values of rradaton and temperature; the second stage nstantaneously estmates the rradaton wth the assumpton that the temperature s constant over one-hour tme ntervals; the thrd stage updates the estmated value of the temperature once every one hour. After estmatng the rradaton and temperature, the voltage correspondng to the MPP s estmated, as well. Then, the reference PV voltage s gven to the power electroncs nterface. The proposed strategy s robust to rapd changes n solar rradaton and load, and t s also nsenstve to ambent temperature varatons. Smulatons studes n PSCAD/EMTDC and Matlab demonstrate the effectveness of the proposed technque. v

5 Acknowledgements Frst and foremost, I thank God for helpng me complete ths work and reach ths stage of lfe. Words are not enough to express my grattude for all hs blessngs. I would lke to thank my supervsor, Dr. Otman Basr, for hs support and gudance over the course of my research and for the dscussons we had that helped brng out ths work. I would also lke to thank Dr. Mohamed-Yaha Dabbagh and Dr. Tarek EL-Fouly for takng the tme to read my thess. My deepest grattude goes to my famly members, especally my mother and father for ther patence and support. I thank all my brothers and ssters as well. v

6 Dedcaton In memory of my grandparents v

7 Table of Contents Author s Declaraton... Abstract... Acknowledgements... v Dedcaton... v Table of Contents... v Lst of Fgures... x Lst of Tables... x Chapter 1 Introducton Photovoltac Systems Overvew Maxmum Power Pont Trackng for Photovoltac Systems... 3 Chapter 2 Components of Grd-Connected Photovoltac Systems Photovoltac Cell Technologes and Modelng Crystallne Slcon PV Cells Mult-Crystallne Slcon PV Cells Thn Flm PV Technologes Photovoltac Cell Equvalent Crcut Model Swtched mode DC-DC Converters Buck Converter Boost Converter Buck-Boost Converter Three Phase Inverters (DC-AC Converters) Maxmum Power Pont Trackng (MPPT) Technques Perturb and Observe Incremental Conductance Fuzzy Logc-based MPPT Neural Networks Chapter 3 Control of Three-Phase Grd Connected PV System Support Vector Regresson Theory Proposed SVR Based Algorthm (MSA) Control of Grd-Interfacng Inverter Chapter 4 Smulaton Results v

8 Chapter 5 Concluson and Future Work Concluson Future Work References Appendx A System Data v

9 Lst of Fgures Fg. 1-1: Autonomous PV system wth batteres... 1 Fg. 1-2: PV hybrd system... 2 Fg. 1-3: Grd-connected PV system... 3 Fg. 1-4: Example for varatons of daly ar temperature at Waterloo, ON, Canada... 4 Fg. 1-5: Example for varatons of daly solar radaton at Waterloo, ON, Canada... 5 Fg. 2-1: Components of a grd connected PV system... 6 Fg. 2-2: Double exponental PV cell model... 9 Fg. 2-3: Smplfed PV cell model Fg. 2-4: I-V characterstc of a Photovoltac module Fg.2-5: Power-Voltage relatonshp of a PV module Fg. 2-6: I-V characterstcs of the PV module under dfferent solar rradaton levels Fg. 2-7: P-V characterstcs of the PV module under dfferent solar rradaton levels Fg. 2-8: I-V characterstcs of the PV module at dfferent surface temperatures Fg. 2-9: P-V characterstcs of the PV module at dfferent surface temperatures Fg. 2-10: Schematc dagram of a buck DC converter Fg. 2-11: Schematc dagram of a DC boost converter Fg. 2-12: Three-phase voltage source nverter (VSI) Fg. 2-13: Flowchart of the perturb and observe algorthm Fg. 2-14: Flowchart of the Incremental conductance MPPT algorthm Fg. 2-15: Neural network for MPPT control Fg.: 3-1 Data transformaton from nput space to kernel space Fg. 3-2: Schematc Dagram of the SVR Fg. 3-3: I-V curves for PV under varable rradaton (G) n W/m2 and temperature (T) n C Fg. 3-4: The flow chart of the proposed MSA Fg. 3-5: The power and control crcut for the DC-DC boost converter Fg. 3-6: Voltage controller block dagram Fg. 3-7: Grd-nterfacng nverter control Fg. 4-1: MSA performance for T=25 o C Fg. 4-2: MSA performance for T=30 o C Fg. 4-3: MSA performance for T=35 o C Fg. 4-4: MSA performance for step change n T from 25 to 35 o C x

10 Lst of Tables Table 2-1: Perturbaton drectons for the P&O algorthm based on output power varatons Table 2-2: Rule based look-up table for fuzzy logc MPPT x

11 Chapter 1 Introducton Photovoltac (PV) systems are used to convert sunlght nto electrcty. They are a safe, relable, low mantenance source of solar electrcty that produces no on-ste polluton or emssons. PV systems ncur few operatng costs and are easy to nstall on most homes. 1.1 Photovoltac Systems Overvew PV systems are categorzed nto three types [1]: autonomous, hybrd and grd-connected. The type of the system wll depend on needs, locaton and budget. Autonomous, or stand-alone, systems are completely ndependent of other power sources. They are usually used to power remote homes, cottages or lodges as well as n applcatons such as remote montorng and water pumpng. In most cases, an autonomous system wll requre batteres for storage. Such systems are partcularly useful and cost-effectve for summer applcatons, when access to a ste s dffcult or costly, or when mantenance needs to be mnmzed. Fg. 1-1: Autonomous PV system wth batteres [1]. 1

12 Fg. 1-2: PV hybrd system [1]. Hybrd systems receve a porton of ther power from one or more addtonal sources. In practce, PV modules are often pared wth a wnd generator or a fuel-fred generator. Such systems usually requre batteres for storage. They are most approprate when energy demand s hgh (n the wnter or year-round), when power must be avalable on demand, or f the budget s lmted. Grd-connected systems allow reducng consumpton from the electrcty grd and, n some nstances, feedng the surplus energy back nto the grd. In some cases, the utlty may gve credt for the energy returned to the grd. Snce power s normally stored n the grd tself, batteres are not necessary unless some form of autonomous power s desrable durng outages. These systems are used n buldngs, homes or cottages already hooked up to the electrcal grd. The man focus n ths thess wll be on these grd-connected PV systems. 2

13 Fg. 1-3: Grd-connected PV system [1]. 1.2 Maxmum Power Pont Trackng for Photovoltac Systems PV modules stll suffers from relatvely low energy converson effcences. Hence, operatng a PV system at ts maxmum power pont s crucal. Therefore, many MPPT methods have been ntroduced n the lterature. Perturb and observe s the most commonly used MPPT method due to ts ease of mplementaton [2], [3]. However, one of the major drawbacks of the P&O method s that under steady state operaton, the output power oscllates around the maxmum power pont. The Incremental Conductance technque was proposed to overcome the dsadvantages of the P&O method [2]. Ths IC technque stops and determnes when the MPP s reached wthout havng to oscllate around ths value. However, t can produce oscllatons and can perform erratcally under rapdly changng atmospherc condtons. Furthermore, the computatonal tme s ncreased due to slowng down of the samplng frequency resultng from the hgher complexty of the algorthm compared to the P&O method. To overcome the demerts of the aforementoned methods, support vector regresson based strateges have been proposed for the estmaton of solar rradaton (for the MPPT) [4]. A sgnfcant advantage of SVR based method s that t can provde the global optmal pont, unlke other methods, for example, artfcal neural networks, whch are prone to multple local mnma. In the publshed lterature, however, researchers have assumed a constant temperature. The assumpton s not plausble 3

14 n practce as the temperature can vary sgnfcantly durng the day. Fg. 1-4 and 1-5 show varatons of the daly temperature and solar rradaton, respectvely, at Waterloo cty, Ontaro, Canada. For each fgure, the darker lne shows the readngs on 10-May-2011, and the lghter lne shows the readngs on 9-May The temperature varaton, n turn, can remarkably affect the effectveness of the MPPT process; the ncluson of temperature measurements n the process wll add to the cost and complexty of the overall PV system. Fg. 1-4: Example for varatons of daly ar temperature at Waterloo, ON, Canada. 4

15 Fg. 1-5: Example for varatons of daly solar radaton at Waterloo, ON, Canada. Ths thess proposes a novel mult-stage algorthm for MPPT n PV systems. The proposed algorthm avods outdoor rradaton and temperature sensors. The proposed MSA conssts of three stages: The frst stage estmates the ntal values of rradaton and temperature; the second stage nstantaneously estmates the rradaton wth the assumpton that the temperature s constant over one-hour tme ntervals; the thrd stage updates the estmated value of the temperature once every one hour. After estmatng the rradaton and temperature, the voltage correspondng to the MPP s estmated, as well. Then, the reference PV voltage s gven to the power electroncs nterface. The proposed strategy s robust to rapd changes n solar rradaton and load, and t s also nsenstve to ambent temperature varatons. Smulatons studes n PSCAD/EMTDC and Matlab demonstrate the effectveness of the proposed technque. 5

16 Chapter 2 Components of Grd-Connected Photovoltac Systems The am of ths chapter s to present a summary of the most mportant buldng blocks n a grd connected PV system. Numerous components are needed to buld a grd connected PV system to carry out the power generaton and converson functons, as shown n fgure 2-1. A PV array converts the lght from the sun nto DC current. A DC-DC converter s connected to the PV array to control ts termnal voltage and to provde the means to mplement an MPPT method. The output power of the array s stored temporarly n large DC lnk capactors to hold the power. A three-phase nverter s then connected to the DC lnk capactors to acheve the power converson of the array DC output power nto AC power sutable for njecton nto the grd. The technque of space vector pulse wdth modulaton (SVPWM) s adopted to control the magntude and phase of the nverter output voltage. A harmoncs flter s located after the nverter to reduce the harmoncs n the output current whch result from the power converson process. An nterfacng transformer s connected after the flter to step up the output AC voltage of the nverter to match the grd voltage level. Protecton relays and crcut breakers are used to solate the PV system when faults occur to prevent damage to the equpment f ther ratngs are exceeded. Fg. 2-1: Components of a grd connected PV system. 6

17 2.1 Photovoltac Cell Technologes and Modelng PV cells are classfed based on the sort of materals used n manufacturng them. Below are some of the common materals used to construct PV cells Crystallne Slcon PV Cells Comprsng 20% of the earth s crust composton, Slcon s consdered the second most abundant element on earth [5]. Slcon exsts n nature n the form of Slcon doxde mnerals lke quartz and slcate based mnerals. It has frst to reach a hgh degree of purty before t can be used for manufacturng sngle crystal PV cells. Hgh grade quartz or slcates are frst treated chemcally to form an ntermedate slcon compound (Lqud trchloroslane SHCl3), whch s then reduced n a reacton wth hydrogen to produce chunks of hghly pure Slcon, about % n purty. After that, these chunks of slcon are melted and formed nto a sngle large crystal of Slcon through a process called the Czochralsk process. The large Slcon crystal s then cut nto thn wafers usng specal cuttng equpment. These wafers are then polshed and doped wth mpurtes to form the requred p-n juncton of the PV cell. Antreflectve coatng materals are added on top of the cell to reduce lght reflectons and allow the cell to better absorb sunlght. A grd of contacts made of slver or alumnum s added to the cell to extract the electrc current generated when t s exposed to lght. The expermental effcency of Sngle crystal slcon cells s about 25% or slghtly hgher under standard test condtons (1000 W/m2 and 25 C). However, commercal PV modules effcency s n the range of 12%-15% [6]. The process of producng PV cells usng ths technology s qute expensve, whch led to development of new technologes that do not suffer from ths drawback. 7

18 2.1.2 Mult-Crystallne Slcon PV Cells In order to avod the hgh cost of producng sngle crystal solar cells, cheaper mult-crystallne cells were developed. As the name mples, mult-crystallne Slcon solar cells do not have a sngle crystal structure. They are rather derved from several smaller crystals that together from the cell. The gran boundares between each crystal reduce the net electrc current that can be generated because of electron recombnaton wth defectve atomc bonds. However, the cost of manufacturng cells usng ths technology s less than what would be n the case of a sngle crystallne cell. The effcency of modules produced usng ths technology ranges from 11%-14% [6] Thn Flm PV Technologes Thn flm PV cells are manufactured through the deposton of several thn layers of atoms or molecules of certan materals on a holdng surface. They have the advantage over ther crystallne Slcon counterparts n ther thckness and weght. They can be 1 to 10 mcrometers thn as compared to 300 mcrometers for Slcon cells [7]. Another advantage s that they can be manufactured usng an automated large area process that further reduces ther cost. Thn flm PV cells do not employ the metal grd requred for carryng current outsde the cell. However, they make use of a thn layer of conductng oxdes to carry the output current to the external crcut. The electrc feld n the p-n juncton of the cell s created between the surface contacts of two dfferent materals, creatng what s called a heterojuncton PV cell. Thn flm PV can be ntegrated on wndows and facades of buldngs because they generate electrcty whle allowng some lght to pass through. Two common thn flm materals are Copper ndum dselende (CIS for short) and Cadmum Tellurde (CdTe). CIS thn flms are characterzed by ther very hgh absorptvty. PV cells that are bult from ths materal are of the heterojuncton type. The top layer can be cadmum sulfde, whle the bottom layer can be gallum to mprove the effcency of the devce. Cadmum Tellurde PV cells are smlar to CIS n ther constructon and manufacturng process. However, the resstvty of p-type CdTe s qute hgh 8

19 therefore ncreasng the nternal losses. Ths ssue can be addressed through the use of ntrnsc CdTe whle usng a layer of znc tellurde between the cell and the back contacts. Effcency for these technologes s about 10-13% [8]. 2.2 Photovoltac Cell Equvalent Crcut Model The equvalent crcut model of a PV cell s requred n order to smulate ts real performance. One of the models proposed n lterature s the double exponental model [6] shown n fgure 2-2. Usng the physcs of p-n junctons, a cell can be modeled as a DC current source n parallel wth two dodes that represent currents escapng due to dffuson and charge recombnaton mechansms. Two resstances, R s and R p, are ncluded to model the contact resstances and the nternal PV cell resstance respectvely. The values of these two resstances can be obtaned from measurements or by usng curve fttng methods based on the I-V characterstc of the cell. The work done n [9] s an example of usng curve fttng technques to approxmate the values of R s and R p. Fg. 2-2: Double exponental PV cell model. 9

20 The relatonshp between the PV cell output current and termnal voltage s governed by: I I I I D1 D2 ph I I I D1 I D2 V IR R q V IRs exp akt qv IRs exp akt p s 1 1 (2.1) where I ph s the PV cell nternal generated photocurrent, I D1 and I D2 are the currents passng through dodes D 1 and D 2, a s the dode dealty factor, k s the Boltzmann constant ( J/K), T s the cell temperature n degrees Kelvn, q s the electron charge ( C), I 01 and I 02 are the reverse saturaton currents of each dode respectvely. Assumng that the current passng n dode D 2 due to charge recombnaton s small enough to be gnored, a smplfed PV cell model can be reached as shown n fgure 2-3. Ths model provdes a good compromse between accuracy and model complexty and has been used n several prevous works [10], [11] and [12]. In ths case, current I D2 can be omtted from (2.1) and the relaton smplfes to: I s I ph I exp 1 q V IR V IRs akt 0 (2.2) Rp Fg. 2-3: Smplfed PV cell model. 10

21 It s obvous that the relatonshp between the PV cell termnal voltage and output current s nonlnear because of the exstence of the exponental term n 2.1 and 2.2. The presence of the p-n semconductor juncton s the cause behnd ths nonlnearty. The result s a unque I-V characterstc for the cell where the current output s constant over a wde range of voltages untl t reaches a certan pont where t starts droppng exponentally. I-V characterstc of a 200 W PV module by Kyocera s shown n fgure 2-4. A PV module s the outcome of connectng several PV cells n seres to order to rase the output voltage. The characterstc has the same shape except for changes n the magntude of the open crcut voltage. Fg. 2-4: I-V characterstc of a Photovoltac module. 11

22 Another mportant relatonshp n PV cells s the power-voltage characterstc. The product of multplyng the current and voltage s evaluated at each pont n the curve to fnd out how much power can be obtaned as voltage changes. The power-voltage relatonshp for the PV module s depcted n fgure 2-5. Intally, power starts ncreasng as voltage ncreases. A certan pont n the curve s reached where maxmum power output can be obtaned; ths pont s therefore referred to as the Maxmum Power Pont. After ths pont, power starts droppng as the termnal voltage ncreases untl t fnally reaches zero at open crcut voltage. Power output from a PV cell s dctated by the magntude of the load resstance, defned by the dvson of the cell voltage overcurrent, n case of fxed loads. If the load mpedance does not equal the value requred to extract maxmum power, then t s possble to use a swtched mode DC converter to do the matchng between the PV cell and the load. The process of changng the PV array termnal voltage externally to extract maxmum power for dfferent loads s known as Maxmum Power Pont Trackng (MPPT). Several technques can be used to perform ths task as wll be explaned n a followng secton of the thess. Fg.2-5: Power-Voltage relatonshp of a PV module. 12

23 The PV cell characterstcs also depend on external factors ncludng temperature and solar rradaton level. To ncorporate these effects nto the model, two addtonal relatons are used. Output current vares wth solar rradaton and temperature through: G I I n KI T (2.3) G where I n s the nomnal PV cell output current (at 25 C and 1000 W/m 2 ), K I s the current/temperature varaton coeffcent (A/ C), ΔT s the varaton from the nomnal temperature (25 C) and G n s the nomnal solar rradaton (1000 W/m 2 ). The value of K I s relatvely small and ths makes the cell output current lnearly dependent on solar radaton level more than temperature. Temperature, however, has a strong effect on the reverse saturaton current, I 0 n 2.2. The followng relaton can be used to model that effect [9]: I 0 I exp, sc, n K qv K I / akt1 oc n where I sc,n s the nomnal short crcut current of the PV cell, V oc,n s the nomnal open crcut voltage, K I and K V are the current and voltage temperature varaton coeffcents, n A/ C and V/ C, respectvely. The effect of solar rradaton and temperature on the characterstcs of the PV module s depcted n fgures 2-6, 2-7, 2-8 and 2-9. To nvestgate the effect of solar rradaton on the currents and voltages of the module, temperature was held constant at 25 C and the resultng I-V and P-V characterstcs were plotted. Fgure 2-6 shows the I-V characterstcs of the module at dfferent rradaton levels of 500, 800 and 1000 W/m2. The P-V characterstcs are shown n fgure 2-7 for the same rradaton levels mentoned. It s notced that output current s drectly proportonal to changes n solar rradaton as expected from the model. Maxmum output power of the module s reduced by half when the solar rradaton drops to 500 W/m2. However, the open crcut voltage does not change 13 V I T n (2.4)

24 sgnfcantly. To fnd out the effect of temperature on the module performance, solar rradaton level was assumed constant at 1000 W/m2 whle allowng temperature to vary between 25 and 80 C. The result s shown n fgures 2-8 and 2-9 for the I-V and P-V characterstcs as temperature was set to 25, 60 and 80 C respectvely. The open crcut voltage of the module decreases as surface temperature ncreases. Current, on the other hand, ncreases slghtly wth temperature. The maxmum power output of the module reduces as the surface temperature rses. Fg. 2-6: I-V characterstcs of the PV module under dfferent solar rradaton levels. 14

25 Fg. 2-7: P-V characterstcs of the PV module under dfferent solar rradaton levels. Fg. 2-8: I-V characterstcs of the PV module at dfferent surface temperatures. 15

26 Fg. 2-9: P-V characterstcs of the PV module at dfferent surface temperatures. 2.3 Swtched Mode DC-DC Converters DC-DC converters are used n a wde varety of applcatons ncludng power supples, where the output voltage should be regulated at a constant value from a fluctuatng power source, to reduce the rpples n the output voltage or acheve multple voltage levels from the same nput voltage. Several topologes exst to ether ncrease or decrease the nput voltage or perform both functons together usng a sngle crcut. The three basc topologes of DC converters are: buck (step down), boost (step up) and the buck-boost converter topologes Buck Converter The schematc dagram of a buck DC converter s shown n fgure It s composed of two man parts: a DC chopper and an output LC flter to reduce the rpples n the resultng output. The output voltage of the converter s less than the nput as determned by the duraton the semconductor swtch 16

27 Fg. 2-10: Schematc dagram of a buck DC converter Q s closed. Under contnuous conducton mode (CCM), the current I L passng through the nductor does not reach zero. The tme ntegral of the nductor voltage over one perod n steady state s equal to zero. From that, the relaton between the nput and output voltages can be obtaned: V V V n out n t V t V t 0 out on on ton t off out off whered dutycycle (2.5) Boost Converter The boost DC converter s used to step up the nput voltage by storng energy n an nductor for a certan tme perod, and then uses ths energy to boost the nput voltage to a hgher value. The crcut dagram for a boost converter s shown n fgure When swtch Q s closed, the nput source charges up the nductor whle dode D s reverse based to provde solaton between the nput and the output of the converter. When the swtch s opened, energy stored n the nductor and the power supply s transferred to the load. The relatonshp between the nput and output voltages s gven by: V V V t n on out n t V on n t t off V off out t off 0 1 (2.6) 1 d 17

28 Fg. 2-11: Schematc dagram of a DC boost converter Buck-Boost Converter Ths converter topology can be used to perform both functons of steppng the nput voltage up or down, but the polarty of the output voltage s opposte to that of the nput. The nput and output voltages of ths confguraton are related through: V V V t n on out n V t t off on t out off 0 d 1 d (2.7) 2.4 Three Phase Inverters (DC-AC Converters) Voltage source nverters (VSI) are manly used to convert a constant DC voltage nto 3-phase AC voltages wth varable magntude and frequency. Fgure 2-12 shows a schematc dagram of a 3 phase VSI. The nverter s composed of sx swtches S1 through S6 wth each phase output connected to the mddle of each nverter leg. Two swtches n each phase are used to construct one leg. The AC output voltage from the nverter s obtaned by controllng the semconductor swtches ON and OFF to generate the desred output. Pulse wdth modulaton (PWM) technques are wdely used to perform ths task. In the smplest form, three reference sgnals are compared to a hgh frequency carrer waveform. The result of that comparson n each leg s used to turn the swtches ON or OFF. Ths technque s referred to as snusodal pulse wdth modulaton (SPWM). It should be noted that the 18

29 swtches n each leg should be operated nterchangeably, n order not to cause a short crcut of the DC supply. Insulated Gate Bpolar Transstors (IGBTs) and power MOSFET devces can be used to mplement the swtches. Each devce vares n ts power ratngs and swtchng speed. IGBTs are well suted for applcatons that requre medum power and swtchng frequency [13]. Fg. 2-12: Three-phase voltage source nverter (VSI). 19

30 2.5 Maxmum Power Pont Trackng (MPPT) Technques MPPT technques are used to control DC converters n order to extract maxmum output power from a PV array under a gven weather condton. The DC converter s contnuously controlled to operate the array at ts maxmum power pont despte possble changes n the load mpedance. Several technques have been proposed n lterature to perform ths task Perturb and Observe Perturb and observe algorthm s a smple technque for maxmum power pont trackng. It s based on controllng the duty cycle (d) of a DC-DC converter to adjust the PV array termnal voltage at the maxmum power pont [2]. The power output of the array s montored every cycle and s compared to ts value before each perturbaton s made. If a change (ether postve or negatve) n the duty cycle of the DC-DC converter causes output power to ncrease, the duty cycle s changed n the same drecton. If t causes the output power to decrease, then t s reversed to the opposte drecton. The algorthm s represented n fgure The performance of the algorthm s affected by the choce of the perturbaton magntude (Δd) of the converter swtchng duty cycle. Large perturbatons cause large output power fluctuatons around the MPP whle small perturbatons slow down the algorthm. Modfcatons to ths technque are publshed n [3], [14] and [15] to mprove performance whle mantanng the basc prncple of operaton. Table 2-1 llustrates the operaton sequence of the algorthm. 20

31 Fg. 2-13: Flowchart of the perturb and observe algorthm. Table 2-1: Perturbaton drectons for the P&O algorthm based on output power varatons Incremental Conductance Ths algorthm explots the fact that the slope of the power-voltage curve of a PV array s equal to zero at the maxmum power pont, as shown n fgure 2-5. The slope s postve n the area to the left of the maxmum power pont and negatve n the area to the rght. Mathematcally, ths can be summarzed as: dp dv 0, at MPP dp dv 0, leftof MPP (2.8) dp dv 0, rghtof MPP Ths can be smplfed usng the followng approxmaton: IV dv I V di dv I V I V dp dv d / (2.9) 21

32 From that, (2.8) can be rewrtten as: I V I V, at MPP I V I V, leftof MPP (2.10) I V I V, rghtof MPP The ncremental conductance algorthm s llustrated n fgure 2-14 where V ref s used as a reference control sgnal for the DC converter [2], [16], [17]. Smlar to the P&O algorthm, the performance of the ncremental conductance MPPT s affected by the ncrement sze of V ref, used here as the control varable. Fg. 2-14: Flowchart of the Incremental conductance MPPT algorthm [2]. 22

33 2.5.3 Fuzzy Logc-based MPPT Fuzzy logc MPPT control s dvded nto three stages: fuzzfcaton, rule based table look-up and defuzzfcaton. Durng the fuzzfcaton process, control varable are converted from a numercal value to a lngustc representaton lke Postve Bg (PB), Postve Small (PS), Zero (Z), Negatve Small (NS) and Negatve Bg (NB). The slope of the P-V curve (S) and the change of slope (ΔS) were used n [18] as the nput control varables to the fuzzfcaton stage. P n S V S 1 Pn n 1 Vn Sn 1 Sn (2.11) where n s the samplng nterval, V and P are the termnal voltage and output power of the PV array respectvely. Next, a look-up table s used to determne the control acton of the converter duty cycle based on the lngustc magntude of the varables. If the PV array s connected to a boost DC-DC converter, then table 2-2 can be used to ssue the control decsons [19]. The look up table s dependent on the DC converter topology beng used. Snce the control command s also n lngustc format, t should be frst converted to a numercal value durng the defuzzfcaton stage. When both the slope and change of slope varables reach zero, the maxmum power pont s reached. Fuzzy logc controllers can perform maxmum power pont trackng effectvely under changng weather condtons. However, there s some dffculty desgnng the look up table and the rules that govern the operaton of the controller for dfferent converter topologes. In addton, thresholds that defne the lngustc varables are to be carefully selected because they can mpact the algorthm performance. An adaptve fuzzy logc controller was proposed n [20] to dynamcally tune these thresholds to acheve mproved performance. 23

34 Table 2-2: Rule based look-up table for fuzzy logc MPPT [19] Neural Networks Neural networks based MPPT s one of the technques suted for mplementaton usng mcrocontrollers [21], [22]. A neural network s composed of three layers: the nput, hdden and output layers. Inputs to a network can be the array termnal voltage and the solar rradaton level or any other measurements needed by the MPPT algorthm. Each node n the network s referred to as a neuron; these neurons are connected together through lnes that are assocated wth certan weghted sums w j. The effectveness of ths MPPT technque s manly determned by the hdden layer and the amount of tranng the network receved. The tranng process mght span several months or years where the network s subjected to varous measurements obtaned from the PV system. Usng ths nformaton, the weghts between the neurons are tuned to generate the requred output whch could be a command to change a DC converter duty cycle. A dsadvantage of ths technque s the lengthy tranng process t needs before the neural network can accurately track the maxmum power pont, n addton to ts dependency on the characterstcs of the PV array to whch t s connected. 24

35 Fg. 2-15: Neural network for MPPT control. 25

36 Chapter 3 Control of Three-Phase Grd Connected PV System Ths chapter covers the control system developed to operate a grd-connected PV system. Frst, an ntroducton to support vector regresson theory s ntroduced. Second, the proposed control of the boost DC-DC converter s explaned where support vector regresson based mult-stage algorthm s adopted n order to track the maxmum power pont for the PV system. Then, an overvew of controllng the grd nterfacng nverter s provded. 3.1 Support Vector Regresson Theory The dea of the support vector regresson technque s based on computaton of the lnear regresson functon n a hgh dmensonal feature space Φ where the nput data x are mapped usng some nonlnear functon. Thus, the SVR problem can be defned as the determnaton of functon f (x) whch approxmates an unknown desred functon and has the followng form [23] f x) w. ( x) b ( (3.1) n where ( ) denotes the nner product, w R, b R, and Φ denotes a non-lnear transformaton from n R space to hgh dmensonal space (see Fg. 3-1). Consderng a set of tranng data n,..., 1 x, y, 1 n where n x R denotes the nput space of the sample and has a correspondng target value y R, and n s the dmenson of the tranng data. To calculate the parameter vector w, the followng cost functon should be mnmzed [23], [24]: Mn 1 2 w 2 C n 1 f ( x ) y (3.2) subject to 26

37 y x y j f(x ) j e 0 -e x Input space j x f(x ) * Kernel space x Fg. 3-1: Data transformaton from nput space to kernel space. where y w( x ) b, 1,2,.., n, * 0 (3.3) s a cost functon, ε s the permssble error, and C s a pre-specfed value that controls the * cost ncurred by tranng errors. Every vector outsde ε-tube s captured n slack varables,, whch are ntroduced to accommodate unpredctable errors on the nput tranng set. The constrants nclude the term, ε, whch allows a margn of error wthout ncurrng any cost. The value of ε can affect the number of support vectors used to construct the regresson functon. As ε s larger, the support vectors selected are fewer. Hence, vales of ε affect the model complexty. Wth reasonable choce of C, a tradeoff between mnmzng tranng errors and mnmzng the model complexty term 2 w s accomplshed. The key dea to satsfy these constrants s to construct a Lagrange functon from the objectve functon and the correspondng constrants by ncludng the lnear restrctons (3.3) nto (3.2) usng * Lagrange multplers assocated to each sample as, 27

38 28 ) ) ( ( ) ) ( ( ) ( 2 1 * 1 2 n b x w y b x w y y x f C w L (3.4) whch s subject to. 0,,, * * Now (3.4) has to be mnmzed wth regard to prmal varables (w, b, and *, ) and maxmzed wth regard to the Lagrange multplers., * Therefore, by makng zero the gradent of L wth respect to the prmal varables, we obtan the followng condtons x w w L ) ( ) ( 0 * (3.5) 0 ) ( 0 * b L (3.6) C L C L 0 0 * * (3.7) If constrants (3.5)-(3.7) are ncluded n the Lagrange functonal (3.4), the optmzaton problem s then obtaned as n n j n j j j y x x K Max 1 * * 1 1, * * ) ( ) ( ) ( ) )( ( 2 1 (3.8) subject to C n, 0, 0, ) ( * 1 * where K represents the kernel matrx

39 K x x ( x) ( x ), (3.9) Equaton (3.1) can be transformed nto the fnal soluton as subject to f ( x) n 1 ( * ) K( x x) b (3.10) * 0 C, 0 C where only the tranng examples whose correspondng Lagrange multplers are nonzero are nvolved n the soluton, whch are called support vectors. Several choces for the kernel are possble to reflect specal propertes of approxmatng functons; the radal base functon (RBF) s used n ths work: K x, x exp 2 x, x 2 where the s called as kernel parameter. Fg. 3-2 shows the schematc dagram of the SVR. Tranng (off-lne) Tranng data (SVs) Testng (on-lne) Runnng data (testng) Input tranng data, C and Read the runnng data x Compute Kernel functon K x, x ) ( j Compute Kernel functon K( x, x ) Compute Lagrange multplers and bas b *, Fnd the functon f(x) f(x) Fg. 3-2: Schematc Dagram of the SVR. 29

40 I (Amp.) 3.2 Proposed SVR Based Algorthm (MSA) As mentoned earler, the output power of the PV system depends on the solar nsolaton (G) and temperature (T). Fgure 3-3 shows I-V characterstcs for a PV array under varable rradaton and temperature. The fgure can be dvded nto two zones: zone A on the left sde where curves do not ntersect wth each other, and zone B on the rght sde where curves ntersect wth each other n several ponts. If the SVR s used to estmate the rradaton and temperature at the ntersectng ponts on the rght sde, the performance of the estmator wll be deterorated. Therefore, we propose a mult-stage SVR based algorthm to solve the aforementoned problem. The proposed algorthm conssts of three stages: The frst stage estmates the ntal values of temperature and rradaton; the second stage nstantaneously estmates the rradaton assumng that the temperature s constant wthn a one-hour tme span; and the thrd stage updates the estmated temperature once every one hour G=200 & T=25 G=200 & T=30 G=200 & T=35 G=300 & T=25 G=300 & T=30 G=300 & T=35 G=400 & T=25 G=400 & T=30 G=400 & T=35 G=500 & T=25 G=500 & T=30 G=500 & T= Zone A Zone B V (volts) Fg. 3-3: I-V curves for PV under varable rradaton (G) n W/m 2 and temperature (T) n C. 30

41 The proposed MSA s composed of three SVR modules (SVR1, SVR2, and SVR3). SVR1 estmates the temperature (T) as a functon of the voltage and current; the tranng samples for SVR1 are taken from zone A; SVR1 s only used when the reference voltage of the boost converter controller V ref s set to equal V selected whch s a voltage exsts n zone A, ts selecton s left to the desgner. SVR2 estmates the rradaton (G) as a functon of the voltage, current, and temperature; the tranng samples for SVR2 are taken from zones A and B assumng that the temperature s constant wthn a one hour tme span. SVR3 module estmates the voltage correspondng to the MPP (V mp ) as a functon of the rradaton and temperature. Then, V ref s set to equal V mp. Every one hour, the reference voltage s set to equal V selected, and SVR1 updates the temperature. Fg. 3-4 shows the flow chart of the proposed MSA, and Fg. 3-5 shows the power crcut and control crcut of the DC- DC boost converter. The control of the grd-nterfacng nverter s explaned n the nest sesson. 31

42 Start Vref V selected No V V 0. 03V pv ref ref Yes T SVR1 f V pv, I SVR2 G f Vpv, I, T SVR3 V mp f G, T Vref V mp No t 1 hour Yes Repeat Fg. 3-4: The flow chart of the proposed MSA. 32

43 I DC-DC Boost Converter Grd- Interfacng Inverter L b Vpv L f Grd PV Array C pv C dc I V pv MSA for MPPT V ref + Voltage Controller - PWM V pv Fg. 3-5: The power and control crcut for the DC-DC boost converter. 3.3 Control of Grd-Interfacng Inverter To acheve the full control of the grd-sde current, the dc-lnk voltage must be boosted to a level hgher than the ampltude of the lne-lne voltage. The power flow of the grd-sde converter s controlled so as to keep the dc-lnk voltage constant. To mantan the dc-lnk voltage constant and to ensure the reactve power flowng nto the grd at null, the grd-sde converter currents are controlled usng the d-q vector control approach. The dc-lnk voltage s controlled to the desred value by usng an IP controller and the change n the DC-lnk voltage represents a change n the q-axs current. A smple voltage control scheme s shown n Fg A current feed-forward control loop s also used here to mprove the dc-lnk voltage response to load dsturbance. For unty power factor, the demand for the d-axs current s zero. Fg. 3-7 shows a control block dagram of the grd-nterfacng nverter. 33

44 Fg. 3-6: Voltage controller block dagram. Fg. 3-7: Grd-nterfacng nverter control. 34

45 Chapter 4 Smulaton Results To verfy the effectveness of the proposed MSA, PSCAD/EMTDC smulatons (nterfaced wth Matlab) have been performed for dfferent temperatures and rradaton levels, where the schematc dagram of the mplemented system was shown n Fg Fgs. 4-1, 4-2, and 4-3 show that the proposed MSA can estmate perfectly the ntal temperature values 25, 30, and 35 o C, respectvely. For each temperature, the rradaton levels are 200, 300, 400, and 500 W/m 2. Frst, the reference voltage V ref s set to equal V selected, whch equals 9 volts, whle SVR1 s actvated to estmate the temperature. Then, the temperature value s kept constant at ts last estmated value before t=0.015 sec when V pv equals V ref. After t=0.015 sec, SVR3 s actvated to estmate the voltage V mp correspondng to the maxmum power, and V ref s set to equal V mp. The Fgures also show that the proposed MSA can estmate the rradaton level and track the MPP wth a hgh accuracy. 35

46 Fg. 4-1: MSA performance for T=25 o C. 36

47 Fg. 4-2: MSA performance for T=30 o C. 37

48 Fg. 4-3: MSA performance for T=35 o C. 38

49 Fg. 4-4 shows the response of the proposed MSA to a step change n the temperature. At t=0.15 sec, the temperature changed from 25 to 35 o C. As a consequence, the percentage error for both rradaton level and output power s ncreased to approxmately 5.2. At t=0.207 sec, V ref s set to equal V selected, and SVR1 s actvated to update the temperature accurately. As a result, the percentage error for both rradaton level and output power s decreased agan to almost zero. 39

50 Fg. 4-4: MSA performance for step change n T from 25 to 35 o C. 40

51 Chapter 5 Concluson and Future Work 5.1 Concluson In ths thess, a novel senseless SVR based mult-stage algorthm (MSA) for MPPT n PV systems s proposed. The proposed algorthm avods usng outdoor rradaton and temperature sensors whch means lower cost. Moreover, t mproves the relablty of the system based on the fact usng less number of sensors mproves the relablty of the system. The proposed MSA conssts of three stages: The frst stage estmates the ntal values of rradaton and temperature; the second stage nstantaneously estmates the rradaton wth the assumpton that the temperature s constant over one-hour tme ntervals; the thrd stage updates the estmated value of the temperature once every one hour. After estmatng the rradaton and temperature, the voltage correspondng to the MPP s estmated, as well. Then, the reference PV voltage s gven to the power electroncs nterface. The proposed strategy s robust to rapd changes n solar rradaton and load, and t s also nsenstve to ambent temperature varatons. Smulatons studes n PSCAD/EMTDC and Matlab demonstrate the effectveness of the proposed technque. 5.2 Future Work As a future work, ncorporaton of the partal shadng effect of PV arrays can be carred on to expand the scope of ths research. Ths problem arses when sunlght s blocked from httng a secton of a PV array drectly whle other sectons contnue ther normal operaton. The estmaton process of the rradaton and temperature of the PV array wll not be accurate and ths wll cause sgnfcant devatons from the maxmum power pont. The control system should be able to detect ths stuaton 41

52 and carry out any requred correctons. Ths control requrement should be carred on close to real tme n order to avod any delays that would degrade the dynamc response of the algorthm. 42

53 References [1] Natural Resources Canada, Photovoltac Systems Introducton, [Onlne]. Avalable: Systems Introducton.pdf [2] T. Esram and P. L. Chapman, "Comparson of Photovoltac Array Maxmum Power Pont Trackng Technques," IEEE Transactons on Energy Converson, vol. 22, pp , [3] N. Fema, G. Petrone, G. Spagnuolo, and M. Vtell, "Optmzaton of perturb and observe maxmum power pont trackng method," IEEE Transactons on Power Electroncs, vol.20, no.4, pp , July [4] Byunggyu Yu; A.G. Abo-Khall, Junghun So, and Gwonjong Yu, Support vector regresson based maxmum power pont trackng for PV grd-connected system, n Proc. IEEE Photovoltac Specalsts Conference (PVSC), Phladelpha, PA, June 2009, pp [5] Masters, Glbert M., Renewable and effcent electrc power systems. John Wley & Sons, [6] J.A. Gow and C. Mannng, "Development of a photovoltac array model for use n powerelectroncs smulaton studes," n Proc. IEE Electrc power applcatons, vol. 146, no. 2, pp , March [7] US Department of Energy (DOE). (2010). Solar Energy Technologes Program. [Onlne]. Avalable: [8] Internatonal Energy Agency (IEA)-PVPS. (2009, Sept.). Trends n photovoltac applcatons. [Onlne]. Avalable: [9] M. G. Vllalva, et al., "Comprehensve Approach to Modelng and Smulaton of Photovoltac Arrays," IEEE Transactons on Power Electroncs, vol. 24, pp , [10] S. B. Kjaer, J. K. Pedersen, and F. Blaabjerg, A revew of sngle-phase grd-connected nverters for photovoltac modules, IEEE Trans. Ind. Appl., vol. 41, no. 5, pp , Sep./Oct

54 [11] I. Atlas and A. Sharaf "A photovoltac array smulaton model for Matlab Smulnk GUI envronment," n nternatonal Conf. on Clean Electrcal Power, pp , May [12] Shengy Lu and R.A. Dougal, "Dynamc multphyscs model for solar array," IEEE Transactons on Energy Converson, vol.17, no.2, pp , Jun [13] Bn Wu, Hgh-Power Converters and AC Drves. John Wley & Sons, [14] Wedong Xao and Dunford, "A modfed adaptve hll clmbng MPPT method for photovoltac power systems," 35th IEEE Annual Power Electroncs Specalsts Conference. PESC 04. vol. 3, no., pp Vol.3, June [15] N.S. D'Souza, Lopes, L. A. C., and Xuejun Lu, "An Intellgent Maxmum Power Pont Tracker Usng Peak Current Control," 36th IEEE Power Electroncs Specalsts Conference, June [16] K.H. Hussen, I. Muta, T. Hoshno, and M. Osakada, "Maxmum photovoltac power trackng: an algorthm for rapdly changng atmospherc condtons," IEE Proceedngs on Generaton, Transmsson and Dstrbuton, vol.142, no.1, pp.59-64, Jan [17] Tae-Yeop Km, Ho-Gyun Ahn, Seung Kyu Park, and Youn-Kyun Lee, "A novel maxmum power pont trackng control for photovoltac power system under rapdly changng solar radaton," IEEE Internatonal Symposum on Industral Electroncs., vol.2, no., pp vol.2, [18] N. Khaehntung, K. Pramotung, B. Tuvrat, and P. Srsuk, "RISC-mcrocontroller bult-n fuzzy logc controller of maxmum power pont trackng for solar-powered lght-flasher applcatons," n Proc. IECON 2004, vol.3, no., pp Vol. 3, 2-6 Nov [19] Chung-Yuen Won, Duk-Heon Km, Se-Chan Km, Won-Sam Km, and Hack-Sung Km, "A new maxmum power pont tracker of photovoltac arrays usng fuzzy controller,"., 25th Annual IEEE Power Electroncs Specalsts Conference, PESC '94, pp vol.1, Jun

55 [20] N. Patcharaprakt and S. Premrudeepreechacharn, "Maxmum power pont trackng usng adaptve fuzzy logc control for grd-connected photovoltac system," IEEE Power Engneerng Socety Wnter Meetng, vol.1, no., pp vol.1, [21] L. Zhang, Yunfe Ba, and A. Al-Amoud, "GA-RBF neural network based maxmum power pont trackng for grd-connected photovoltac systems," Internatonal Conference on Power Electroncs, Machnes and Drves, vol., no., pp , 4-8 June [22] T. Hyama and S. Kouzuma, T. Imakubo, "Identfcaton of optmal operatng pont of PV modules usng neural network for real tme maxmum power trackng control," IEEE Transactons on Energy Converson, vol.10, no.2, pp , Jun [23] A. J. Smola and B. Schölkopf, A tutoral on support vector regresson, Statst. Comput., vol. 14, pp , [24] K. R. Muller, A. Smola, G. Ratsch, B. Scholkopf, J. Kohlmorgen, and V. Vapnk, "Predctng tme seres wth support vector machnes," n Proc. ICANN. New York: Sprnger-Verlag, 1997, vol. 1327, pp

56 Appendx A System Data Parameters n Fg. 3-5 C pv L b C dc L f V L-G f Value 500 [µf] [H] 2000 [µf] [H] 208 [V] 60 [Hertz] 46