ANN Methods in photovoltaic system MPPT controller ... · Maximum Power Point Tracking (MPPT) Control system. MPPT controller for uniform condition MPPT controller works based on

ANN Methods in photovoltaic system

MPPT controller (uniform & non-uniform irradiance conditions)Performance of TFFN, RBF and ANFIS networks

for different technology of PV modules

MPPT controller for uniform conditionSyafaruddin, Engin Karatepe, Takashi Hiyama: ‘Polar Coordinated Fuzzy Controller based Real-Time Maximum Power

Point Control of Photovoltaic System’, Renewable Energy, Vol. 34, No. 12, pp. 2597-2606, December, 2009.

Energy DemandEnergy Demand

Environmental ProblemsEnvironmental Problems

Distributed GenerationDistributed Generation(Research & Development)(Research & Development)

Photovoltaic SystemPhotovoltaic System

Other Advantages of PV System:Other Advantages of PV System:*Unlimited fuel supply*Unlimited fuel supply*Clean *Clean -- no pollution or emissions & waterno pollution or emissions & water*Silent*Silent*Minimal visual impact*Minimal visual impact*Low maintenance*Low maintenance*Easily expanded*Easily expanded*Generates at load (Distributed Generation)*Generates at load (Distributed Generation)

Abundant fuel & materialsAbundant fuel & materials Source: Global Energy Network Institute (GENI)

Applications:*power supplies for satellite communications *large power stations feeding electricity into the grid

MPPT controller for uniform condition

Challenges to the widespread Challenges to the widespread use of photovoltaic system:use of photovoltaic system:

high cost of module high cost of module materials and encapsulationmaterials and encapsulationPV power conditioning PV power conditioning system (critical issue for system (critical issue for efficient PV system)efficient PV system)intermittency output intermittency output characteristicscharacteristicsnonnon--linear characteristiclinear characteristic

Lower EfficiencyLower Efficiency

PV system should be PV system should be operated optimallyoperated optimally

The state of the art techniques to The state of the art techniques to track the maximum available track the maximum available output power of PV systemsoutput power of PV systems

Maximum Power Point Maximum Power Point Tracking (MPPT) Control Tracking (MPPT) Control

systemsystem


MPPT controller works based on certain developed algorithms:MPPT controller works based on certain developed algorithms:Perturbation and observation (P&O)Perturbation and observation (P&O)

Incremental conductance Incremental conductance Fractional open circuit voltage methods Fractional open circuit voltage methods

P&O and incremental conductance methods:widely used in PV applicationswidely used in PV applications

Combine to improve the tracking accuracyCombine to improve the tracking accuracy the efficiency of tracking techniques strongly depends on iteration step sizethe efficiency of tracking techniques strongly depends on iteration step size

Fractional open circuit voltage method:The optimal point The optimal point a constant value k=MPP Voltage/Va constant value k=MPP Voltage/Vococ

Conventional methods weaknessesConventional methods weaknesses!!!!!!


The Weaknesses of Conventional MPPT ControllersThe Weaknesses of Conventional MPPT Controllers

P&O and incremental conductance methods:P&O and incremental conductance methods:

continue oscillation causes power lossesalgorithms to the fast dynamic responsevery difficult to get an analytical expression in different solar cells technologies

Fractional open circuit voltage method:

MPP voltage -Vs- irradiance might not be same for different solar cell technologies Therefore, it is impossible to determine the optimum voltage by using only one linear function of the open circuit voltage

ANN-FUZZY LOGIC SCHEMES BASED MPPT CONTROLLER

Mathematical Model from Sandia National Laboratory

Syafaruddin, Engin Karatepe, Takashi Hiyama: ‘ANN based Real-Time Estimation of Power Generation of Different PV Module Types’, IEEJ Transaction on Power and Energy, Vol. 129, No. 6, pp. 783-790, June, 2009.

I-V and P-V characteristics of PV modules (E=100-1000W/m2, Tc=50oC)

PV Modules TechnologyPV Modules Technology

Siemens SMSiemens SM--55 PV (c55 PV (c--Si)Si)high efficiency output power high efficiency output power under reduced light condition by under reduced light condition by using pyramidal textured surfaceusing pyramidal textured surface

First Solar FSFirst Solar FS--50 50 (Thin film CdTe) (Thin film CdTe) thin layers of compound thin layers of compound semiconductor material with low semiconductor material with low temperature coefficients which temperature coefficients which provides for cost effective and provides for cost effective and greater energy productiongreater energy production

USSC USUSSC US--21 (3j a21 (3j a--Si)Si)composed of encapsulated composed of encapsulated polymer inside a rigid anodized polymer inside a rigid anodized aluminum framealuminum frame

ANN-FUZZY LOGIC CONTROL SCHEMES BASED MPPT CONTROLLER

Intelligent System(ANN)

modelingmodelingidentificationidentificationoptimizationoptimizationforecasting forecasting control control

Solve the engineering problems:Solve the engineering problems:••symbolic reasoningsymbolic reasoning••flexibility flexibility ••explanation capabilitiesexplanation capabilities••capable to handle and learn the noncapable to handle and learn the non--linear, linear, large, complex and even incomplete data large, complex and even incomplete data patternspatterns

Complex SystemComplex System(different fields of application)(different fields of application)

Compared standard linear model for Compared standard linear model for optimization methods:optimization methods: compact solution for multicompact solution for multi--variable variable problemsproblems not require the knowledge of internal not require the knowledge of internal system parameters system parameters only training process is required and the only training process is required and the output parameters are directly determined output parameters are directly determined without solving any nonwithout solving any non--linear mathematical linear mathematical equations or statistical assumptions as in the equations or statistical assumptions as in the conventional optimization methods conventional optimization methods

MotivationMotivation--11


three layer feed-forward neural network The important steps of ANN The important steps of ANN method:method:Selection of training data setSelection of training data set

Training processTraining processValidationValidation

Selection of training data set:Selection of training data set:••To set the functions between input and To set the functions between input and output through the hidden layers output through the hidden layers (training (training data set should be selected to cover the entire data set should be selected to cover the entire region where the network is expected to region where the network is expected to operate) operate) ••Training data pattern Training data pattern [[VVmpmp,P,Pmpmp]=f(]=f(E,TE,Tcc); ); following the mathematical modelfollowing the mathematical model••Total data (228 sets) for operating conditions Total data (228 sets) for operating conditions between 15between 15--6565ooC and 100C and 100--1000W/m1000W/m22

ANN-FUZZY LOGIC CONTROL SCHEMES BASED MPPT

CONTROLLER )(11)( kIi ie

kO

j

jiji kOkwkI )()()(

2

1)()(

N

kkOktSSE

In the hidden and output layers, the In the hidden and output layers, the sigmoid function is utilized for the inputsigmoid function is utilized for the input--output characteristics of the nodes. For output characteristics of the nodes. For each node each node ii ; the output ; the output OOii(k)(k)

IIii(k)(k) : the input signal to node : the input signal to node ii at the at the kk--th th sampling and this is given by the weighted sampling and this is given by the weighted sum of the input nodessum of the input nodes

wwijij : the connection weight from node : the connection weight from node jj to node to node iiOOjj(k):(k): the output from nodethe output from node jj

During the training, the connection weights During the training, the connection weights wwijij are tuned recursively until the best fit is are tuned recursively until the best fit is achieved for the inputachieved for the input––output patterns output patterns based on the minimum value of the sum of based on the minimum value of the sum of the squared errors (SSE) the squared errors (SSE)

NN is the total number of training patterns, is the total number of training patterns, t(k)t(k) and and O(k)O(k) are the are the kk--th output target and estimated th output target and estimated values of MPPs voltagevalues of MPPs voltage--power power

During the training process:During the training process:learning rate = 0.2learning rate = 0.2

momentum rate = 0.85momentum rate = 0.85

TFNN: TFNN:

(average FF: 0.735241) (average FF: 0.735241)



Siemens SMSiemens SM--55; c55; c--SiSi

First Solar FSFirst Solar FS--50; 50; Thin Film CdTeThin Film CdTe

Uni Solar USUni Solar US--21; 3j a21; 3j a--SiSi


0

0.01

0.02

0.03

0.04

0.05

0.06

2 3 4 5 6 7 8 9 10

number of hidden nodes

SS

E

SM-55 FS-50 US-21

ANN Training ResultsANN Training Results

Number of hidden nodes based on the minimum of sum of the squared error (SSE)

Fill factor of PV modulesFill factor of PV modulesEE=100=100--1000W/m1000W/m22 and and TTcc=10=10--6060ooCC

The cells made of crystalline silicon generally have higher fill The cells made of crystalline silicon generally have higher fill factor than those of the others factor than those of the others “more hidden nodes are required to accurately represent the “more hidden nodes are required to accurately represent the characteristics of crystalline silicon cells” characteristics of crystalline silicon cells”

PV ModulesPerformance Index

JV JPSM-55 0.1674 0.1091

FS-50 0.6774 0.2057

US-21 0.2071 0.1501


Verification of ANN training results Verification of ANN training results during 12 hours of simulation time during 12 hours of simulation time

dtVVJ mpdcV2)*(

dtPPJ mpdcP2)*(

Smallest values of both Smallest values of both JJVV and and JJPP are expected in this case that indicates how close are expected in this case that indicates how close the values between the optimum and the target the values between the optimum and the target

VVmpmp & & PPmpmp : : MPP points from PMPP points from P--V curve (target in training process)V curve (target in training process)

VVdcdc* * and and PPdcdc*: *: optimum voltage and power from ANNoptimum voltage and power from ANN

“the prediction of global MPP voltage as a reference signal for controller is one of the

solutions to improve the stability of the MPPT controller”


MPP voltage as reference MPP voltage as reference To establish the control To establish the control signal in realsignal in real--time for keeping time for keeping the voltage of PV systems at the voltage of PV systems at optimum operating pointoptimum operating pointApplied for the first time in Applied for the first time in the MPPT systemthe MPPT system

Advantages:*no required accurate description of the system to be controlled*no wide parameter variations with respect to the standard regulators

Polar Coordinated Fuzzy ControllerPolar Coordinated Fuzzy Controller Similar fuzzy logic rules in the PSS application


Rule base:Rule base:(Fuzzy rules assignment table)(Fuzzy rules assignment table)

)(),(_)( keAkeIntkZ s

Int_e(k) and e(k) are the state of integral of error and error, respectively and As is the scaling factor of the error

PSS: Int_e(k)=speed deviatione(k)= acceleration

e(k) =Vdc*(k) – Vdc(k)

Three important stages as Rule base, Fuzzification and DefuzzificationThree important stages as Rule base, Fuzzification and Defuzzification

Phase plane of fuzzy logic control Phase plane of fuzzy logic control with polar information with polar information

Increase in Increase in VVdcdc

Reduction in Reduction in VVdcdc

FuzzificationFuzzification stage: stage: numerical variables are transformed into linguistic variablesnumerical variables are transformed into linguistic variables

)(_)(tan)( 1keInt

keAk s

22 )())(_()( keAkeIntkD s

r

rr

DkDfor

DkDforD

kDkDG

)(0.1

)()())((

Angle membership function

Radius membership functionTuning parameters of Tuning parameters of AAss and and DDrr are specified are specified at 5.0 and 0.35,respectively at 5.0 and 0.35,respectively

PhasePhaseplaneplane

Defuzzification stage: linguistic variables are converted back into numerical variableslinguistic variables are converted back into numerical variables(weighted averaging weighted averaging defuzzificationdefuzzification algorithmalgorithm)

The voltage control signal (UCNV) will be able to maintain the operating voltage of PV system to its optimum voltage

max)).((.))(())(())(())(()( UkDG

kPkNkPkNkUCNV

Numerical variables:Numerical variables: angle & radius angle & radius Linguistic variables:Linguistic variables: angleangle:: increaseincrease and reduction in the and reduction in the operating voltage; operating voltage; radiusradius: : near and far from target pointnear and far from target point

MPPT controller for non-uniform conditionsSyafaruddin, Engin Karatepe, Takashi Hiyama: ‘Artificial Neural Network - Polar Coordinated Fuzzy Controller

based Maximum Power Point Tracking Control under Partially Shaded Conditions’, IET Renewable Power Generation, Vol. 3, No. 2, pp. 239-253, June, 2009

Another recent issue: Partially Shaded ConditionsAnother recent issue: Partially Shaded Conditions(inevitable in PV system practice)(inevitable in PV system practice)

“A condition when some parts of module or array receive less intensity of sunlight”“A condition when some parts of module or array receive less intensity of sunlight”

Utility poles Trees

Chimneys or parts of other buildings

Dirt on the module’s top surface

Under the partially shaded conditions:Under the partially shaded conditions:the shaded cells or modules will force the output the shaded cells or modules will force the output

current of noncurrent of non--shaded partsshaded partsto be low following the output current of to be low following the output current of

the shaded ones;the shaded ones;Consequently, the output power of PV array is Consequently, the output power of PV array is

significantly reduced significantly reduced

Significant efforts to solve Significant efforts to solve this kind of mismatching lossesthis kind of mismatching losses

the novel techniquesthe novel techniquesto minimize the losses of to minimize the losses of

partial shading require more partial shading require more additional sensorsadditional sensors, ,

auxiliary algorithmsauxiliary algorithms and and power electronics unitspower electronics units

APPLICATION FOR PARTIALLY SHADED CONDITIONSAPPLICATION FOR PARTIALLY SHADED CONDITIONS

The output characteristic of a photovoltaic The output characteristic of a photovoltaic (PV) array is changed depending on:(PV) array is changed depending on:solarsolar irradiance,irradiance, temperature,temperature, mismatchedmismatchedcells,cells, arrayarray configurationconfiguration andandpartiallypartially shadedshaded conditionsconditions

Under the Partially Shaded Conditions:Under the Partially Shaded Conditions:The (The (II--VV) and () and (PP--VV) characteristics get ) characteristics get more complicated with multimore complicated with multi--local local maximum power point (MPP)maximum power point (MPP)(tracking the MPP power is difficult) (tracking the MPP power is difficult)

MPPT Control must be developed:MPPT Control must be developed:To achieve highest possible performance of PV conversion and cover the To achieve highest possible performance of PV conversion and cover the

high cost of solar cellshigh cost of solar cells

Conventional controllers may not Conventional controllers may not guarantee fully working under this guarantee fully working under this condition because they can not condition because they can not distinguish between the global and distinguish between the global and local peaks since local MPP shows local peaks since local MPP shows the same typical characteristics as the same typical characteristics as global MPP & they converge to a global MPP & they converge to a local MPPlocal MPP

Thus, the proper amount of power generation can not be utilized by using conventional algorithms when partially shaded occurs


Significant efforts to reduce this kind of mismatching losses: Significant efforts to reduce this kind of mismatching losses:

Shimizu et alShimizu et al proposed the operation voltage control circuit where dcproposed the operation voltage control circuit where dc--dc converter is dc converter is installed in each module. installed in each module.

Mishima and OhnishiMishima and Ohnishi proposed a system with simplification of output power control of array proposed a system with simplification of output power control of array on a PV string basis.on a PV string basis.

Kobayashi et al. and Irisawa et al.Kobayashi et al. and Irisawa et al. proposed two sequential stages of MPPT control system. proposed two sequential stages of MPPT control system.

All the efforts for overcoming the partially shaded problems show that All the efforts for overcoming the partially shaded problems show that advanced MPPT systemsadvanced MPPT systems should be developed and should be developed and the investigation of the investigation of finding optimal solutionsfinding optimal solutions should be increased to get more reliable PV should be increased to get more reliable PV systemsystem


ModifiedModifiedANNANN

ModifiedModifiedPV system PV system (PV array; (PV array;

different size & different size & configuration)configuration)

Siemens SM-55 PV Modules36 series monocrystalline Si cells

Maximum power Open circuit voltage Short circuit current Voltage at maximum power point Current at maximum power point STC: AM 1.5, 1000W/m2, 25oC

55 W21.7 V3.45 A17.4 V3.15 A


ModifiedModifiedPV systemPV system

Specification of SM 55 PV module Specification of SM 55 PV module

021 loadVVV

01)(exp)(

)(1)()(18)((exp)()(

bdbd

isbd

p

siisiisphi kTn

VqIiR

iRIVikTiniRIVqiIiII

021 II

For i=1,2For i=1,2

q = the electric charge (1.6x10-19C)k = Boltzmann constant (1.38x10-23 J/K)Bypass diode: Isbd=1.6e-9A, nbd=1.0 and Tbd=35oC

Modular Model PV Array size 3x3 (0.5kW), 20x3 (3.3kW) & PV connection: SP, BL, TCT


Arr

ay P

ower

[ W

]

••To prevent the “hot spot problem”To prevent the “hot spot problem”••To give an opportunity to get more power under To give an opportunity to get more power under shading conditionshading conditionHowever, The shaded parts can force the However, The shaded parts can force the optimum voltage to be abnormally low & Multiple optimum voltage to be abnormally low & Multiple peakspeaks

Roles of bypass diode:

No shading:Both bypass diode are in reverse biased voltageCell-1 non-shaded, Cell-2 shaded:Dbd1 reverse biased; Dbd2 forward biasedOr vice versa

APPLICATION FOR APPLICATION FOR PARTIALLY SHADED PARTIALLY SHADED

CONDITIONSCONDITIONS

Modified ANN Modified ANN input signals for input signals for TFFNTFFN

For 3x3 PV array: 4 input signals of EWith dimension 2x2

PV array size of PV array size of AA(m,n)(m,n) and the and the input signals of average input signals of average irradiance on four adjacent PV irradiance on four adjacent PV modules of modules of EE(r,s)(r,s)

For r = 1: mFor r = 1: m--1 and s=1: n1 and s=1: n--1,1,)(25.0),( )1,1(),1()1,(),( srAsrAsrAsrA EEEExsrE

For 20x3 PV array: 38 input signals of EWith dimension 19x2


TFFN for 3x3 PV arrayTFFN for 3x3 PV array

)(25.0)1,1( )2,2()1,2()2,1()1,1( AAAA EEEExE

)(25.0)2,1( )3,2()2,2()3,1()2,1( AAAA EEEExE

)(25.0)1,2( )2,3()1,3()2,2()1,2( AAAA EEEExE

)(25.0)2,2( )3,3()2,3()3,2()2,2( AAAA EEEExE


Base Irradiance [W/m2]

Pre-determined shading in W/m2on selected modules

2003004005006007008009001000

100100, 200100, 200, 300100, 200, 300, 400100, 200, 300, 400, 500100, 200, 300, 400, 500, 600100, 200, 300, 400, 500, 600, 700100, 200, 300, 400, 500, 600, 700, 800100, 200, 300, 400, 500, 600, 700, 800, 900

Training Patterns:Training Patterns: maximum power point (MPP) voltage maximum power point (MPP) voltage ((VVmpmp) and power () and power (PPmpmp)) are observed through are observed through PP--VV curve curve for different shading patterns (as shown in the table) for different shading patterns (as shown in the table)

For 3x3 PV array:SSE= 0.0253 nh= 12For 20x3 PV array:SSE= 0.0335 nh= 44

Learning rate = 0.2Learning rate = 0.2Momentum rate = 0.85 Momentum rate = 0.85

Similar ANN Structure Similar ANN Structure TFFNTFFN

ANN Training Results:ANN Training Results:


Irradiance PV arraynon-shaded 25%-shaded 50%-shaded 75%-shaded

JV JP JV JP JV JP JV JP

Slow changes

3x3

SP 0.247 5.975 0.509 16.303 1.395 19.628 9.076 19.191

BL 0.247 5.975 0.266 19.533 0.550 12.220 0.951 10.573

TCT 0.247 5.975 0.269 5.757 0.463 9.876 0.448 4.709

Rapid changes

SP 1.029 5.295 4.158 14.945 8.031 50.693 11.098 49.350

BL 1.029 5.295 2.458 15.659 6.180 19.025 7.335 39.696

TCT 1.029 5.295 1.250 6.280 4.185 10.148 2.691 10.962

Slow changes

20x3

SP 0.338 7.853 0.618 15.306 3.467 22.346 11.257 22.346

BL 0.338 7.853 0.457 20.654 1.457 13.578 1.897 11.254

TCT 0.338 7.853 0.358 9.798 2.437 11.259 1.574 8.875

Rapid changes

SP 1.463 8.014 5.378 16.403 10.025 45.078 13.236 47.048

BL 1.463 8.014 4.864 18.876 8.725 21.134 10.046 40.136

TCT 1.463 8.014 2.253 10.935 5.276 12.564 4.504 12.056

ANN Output VerificationANN Output Verification Performance Index: Performance Index: JJvv and and JJpp::during 12 hours of simulation time during 12 hours of simulation time

dtVVJ mpdcV2)*(

dtPPJ mpdcP2)*(

DEVELOPMENT OF REALDEVELOPMENT OF REAL--TIME SIMULATORTIME SIMULATOR

Main reasons of using realMain reasons of using real--time simulator:time simulator:“Various possible input scenarios”“Various possible input scenarios” very very difficult to examine the behavior of proposed difficult to examine the behavior of proposed system in the realsystem in the real--system system

Light intensity is Measured in Lux Light intensity is Measured in Lux (1 Lux =0.0161028 W/m(1 Lux =0.0161028 W/m22))

Time accelerated mode:1 sec in simulation = 1 hour in real practice

RealReal--time simulator based dSPACE realtime simulator based dSPACE real--time interface systemtime interface system

Advantages:Advantages:••To overcome field testing (costly & time To overcome field testing (costly & time consuming)consuming)••“What“What--if scenarios” can be simply simulatedif scenarios” can be simply simulated••To allow increasing experience of the PV To allow increasing experience of the PV system behavior as well as in actual system system behavior as well as in actual system

Virtual PV system with MPPT controllerVirtual PV system with MPPT controller

SIMULATION RESULTSSIMULATION RESULTSRealReal--time simulation results in different time simulation results in different input scenarios on PV modulesinput scenarios on PV modules

Irradiance level from 300W/mIrradiance level from 300W/m22 to 800W/mto 800W/m22 at constant at constant TTcc=45=45ooC applied C applied for SMfor SM--55 type PV module 55 type PV module

14

14.4

14.8

15.2

15.6

6

6.45 7.

3

8.15 9

9.45

10.3

11.2 12

12.5

13.3

14.2 15

15.5

16.3

17.2 18

time (sec)

DC

Vol

tage

(V)

Vdc Vdc*

14

14.4

14.8

15.2

15.6

6

6.45 7.

3

8.15 9

9.45

10.3

11.2 12

12.5

13.3

14.2 15

15.5

16.3

17.2 18

time (sec)

DC

Vol

tage

(V)

Vdc Vdc*

Step changeStep change Slow shadingSlow shading

-0.1

0

0.1

0.2

0.3

0.4

Con

trol

Sig

nal,

U CN

V (V

)

-0.4

-0.2

0

0.2

0.4

Con

trol

Sig

nal,

U CN

V (V

)

14

14.4

14.8

15.2

15.6

6

6.45 7.

3

8.15 9

9.45

10.3

11.2 12

12.5

13.3

14.2 15

15.5

16.3

17.2 18

time (sec)

DC

Vol

tage

(V)

Vdc Vdc*

Quick shadingQuick shading

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

Con

trol

Sig

nal,

U CN

V (V

)

“The results confirm that the “The results confirm that the proposed technique is robust proposed technique is robust and insensitive to changes in and insensitive to changes in these weather conditions”these weather conditions”

Daily pattern of EDaily pattern of E

SIMULATION RESULTSSIMULATION RESULTS

Slow changes in irradianceSlow changes in irradiance‘clear sky measurement’‘clear sky measurement’

(6am(6am--6pm) 6pm)

Quick changes in irradianceQuick changes in irradiance‘cloudy sky measurement’‘cloudy sky measurement’

(6am(6am--6pm) 6pm)

SIMULATION RESULTSSIMULATION RESULTSDC voltage, control signal DC power DC voltage, control signal DC power

for SMfor SM--55 PV module 55 PV module

12

14

16

186

6.45 7.

3

8.15 9

9.45

10.3

11.2 12

12.5

13.3

14.2 15

15.5

16.3

17.2 18

time (hours)

DC

Vol

tage

(V)

Vdc Vdc*

0

10

20

30

40

50

66.4

5 7.3 8.15 9

9.45

10.3

11.15 12

12.45 13.3

14.15 15

15.45 16.3

17.15 18

time (hours)

DC

Pow

er (W

)

Pdc Pdc*

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

time (hours)

Con

trol

Sig

nal,

UCN

V (V

)

12

14

16

186

6.45 7.

3

8.15 9

9.45

10.3

11.2 12

12.5

13.3

14.2 15

15.5

16.3

17.2 18

time (hours)

DC

Vol

tage

(V)

Vdc Vdc*

0

10

20

30

40

50

60

6

6.45 7.

3

8.15 9

9.45

10.3

11.2 12

12.5

13.3

14.2 15

15.5

16.3

17.2 18

time (hours)

DC

Pow

er (W

)

Pdc Pdc*

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

time (hours)

Con

trol

Sig

nal,

UCN

V (V

)


The accuracy between the The accuracy between the dcdc voltage controller output (voltage controller output (VVdcdc) and the optimum ) and the optimum voltage from the ANN (voltage from the ANN (VVdcdc**) is measured by ARE & RE) is measured by ARE & RE

Average relative error (Average relative error (AREARE):): %100**1

i dc

dcdc

VVV

KARE

Relative error (Relative error (RERE) to the estimated optimum voltage:) to the estimated optimum voltage: %100*

*

idc

idcdc

tV

tVVRE

IrradianceSM-55 FS-50 US-21

ARE(%) RE(%) ARE(%) RE(%) ARE(%) RE(%)

Step change 0.167 0.115 0.895 0.542 0.410 0.121

Slow shading 0.251 0.165 0.966 0.678 0.590 0.194

Quick shading 0.327 0.127 1.283 0.735 0.733 0.212

Slow changes 1.651 0.753 2.845 0.947 2.054 0.816

Rapid changes 1.837 0.925 3.086 1.134 2.135 0.962


Comparison of Control Performance with the ProportionalComparison of Control Performance with the Proportional--Integral (PI) controllerIntegral (PI) controller

Proportional gain (Proportional gain (KKpp) =1.5) =1.5Integral gain (Integral gain (KKii) = 0.75 ) = 0.75

The reason for comparing our proposed controller:The reason for comparing our proposed controller:to show the robustness, stability and accuracy over to show the robustness, stability and accuracy over the conventional PI controllerthe conventional PI controller

IrradianceSM-55 FS-50 US-21

PI FL PI FL PI FL

Step change 0.0666 0.0109 5.278 2.587 0.7102 0.1205

Slow shading 0.2207 0.0498 6.678 3.384 0.5897 0.0942

Quick shading 0.2268 0.1067 8.997 4.605 0.7331 0.2121

Slow changes 3.7040 1.0880 14.812 5.932 5.0630 0.7665

Rapid changes 3.9245 1.1750 15.735 6.237 5.1234 0.7883 dtVVJ dcdc

2)*(

Performance index of both controllers

during 12 hours of during 12 hours of simulation timesimulation time

SM-55 PV module voltage and current

FS-50 PV module voltage and current

Step changes in Irradiance & cell temperature

Daily condition of irradiance & cell temp.

Syafaruddin, Engin Karatepe, Takashi Hiyama: ‘Development of Real-Time Simulator based on Intelligent Techniques for Maximum Power Point Controller of PV Modules’, The International Journal of Innovative Computing, Information and Control (IJICIC), Vol. 6, No. 4, pp.1623-1642, April, 2010.

Other results RBF methods:


RealReal--time simulation results under partially shaded conditionstime simulation results under partially shaded conditions

Shading patterns of irradiance level Shading patterns of irradiance level

25

30

35

40

45

50

6

6.4

7.2 8

8.4

9.2 10

10.4

11.2 12

12.4

13.2 14

14.4

15.2 16

16.4

17.2 18

time (hours)

DC

Vol

tage

(V)

Vdc Vdc* Vdc_75% Vdc*_75%

Vdc_50% Vdc*_50% Vdc_25% Vdc*_25%

0

100

200

300

400

500

6

6.4

7.2 8

8.4

9.2 10

10.4

11.2 12

12.4

13.2 14

14.4

15.2 16

16.4

17.2 18

time (hours)

DC

Pow

er (W

)

Pdc Pdc* Pdc_75% Pdc*_75%

Pdc_50% Pdc*_50% Pdc_25% Pdc*_25%


25

30

35

40

45

50

6

6.4

7.2 8

8.4

9.2 10

10.4

11.2 12

12.4

13.2 14

14.4

15.2 16

16.4

17.2 18

time (hours)

DC

Vol

tage

(V)

Vdc Vdc* Vdc_75% Vdc*_75%

Vdc_50% Vdc*_50% Vdc_25% Vdc*_25%

0

100

200

300

400

500

6

6.4

7.2 8

8.4

9.2 10

10.4

11.2 12

12.4

13.2 14

14.4

15.2 16

16.4

17.2 18

time (hours)

DC

Pow

er (W

)

Pdc Pdc* Pdc_75% Pdc*_75%

Pdc_50% Pdc*_50% Pdc_25% Pdc*_25%

-1.5

-1

-0.5

0

0.5

1

1.5

time (hours)

Con

trol

Sig

nal,

UCN

V (V

)

non-shaded 75%-shaded 50%-shaded 25%-shaded

-1.5

-1

-0.5

0

0.5

1

1.5

time (hours)

Con

trol

Sig

nal,

U CN

V (V

)

non-shaded 75%-shaded 50%-shaded 25%-shaded

The control responses are verified through an performance indexThe control responses are verified through an performance indexduring 12 hours of simulation timeduring 12 hours of simulation time

Irradiance PV array non-shaded 25%-shaded 50%-shaded 75%-shaded

Slow changes

SP 0.493 1.131 2.406 16.501

BL 0.493 0.592 0.948 1.729

TCT 0.493 0.597 0.798 0.814

Rapid changes

SP 2.057 9.239 13.846 20.178

BL 2.057 5.462 10.655 13.337

TCT 2.057 2.778 7.216 4.893

For 3x3 PV array:For 3x3 PV array:


dtVVJ dcdc2)*(


time (h)

Shading patterns on modules area (W/m2)

SP BL TCT

P & O Proposed Controller P & O Proposed Controller P & O Proposed Controller

A B C D Vdc (V) Pdc (W) Vdc (V) Pdc (W) Vdc (V) Pdc (W) Vdc (V) Pdc (W) Vdc (V) Pdc (W) Vdc (V) Pdc (W)

10.00 250 150 250 750 320.60 485.66 223.32 546.33 320.60 486.69 227.74 554.85 320.60 486.84 227.74 555.47

10.30 150 500 750 750 340.50 514.65 236.58 1179.40 340.50 514.66 236.58 1180.60 340.50 514.67 236.58 1181.00

11.00 500 750 1000 1000 333.87 1683.30 234.37 1747.30 333.87 1683.30 234.37 1748.50 333.87 1683.40 234.37 1748.96

11.30 250 500 750 1000 338.29 854.30 238.79 1199.80 338.29 854.38 238.79 1201.50 338.29 854.43 238.79 1201.90

12.00 500 250 500 750 331.66 835.98 223.32 1089.00 331.66 837.49 225.53 1095.30 331.66 837.73 225.53 1103.00

12.30 150 500 150 500 313.97 465.73 139.30 663.79 313.97 465.73 139.30 663.79 313.97 465.73 139.30 663.79

1.00 500 750 150 250 329.45 499.30 150.35 737.26 329.45 499.33 148.14 728.20 329.45 499.35 148.14 727.79

1.30 750 500 250 500 331.66 838.50 227.74 1107.30 331.66 838.50 227.74 1107.30 331.66 838.50 227.74 1107.30

2.00 1000 750 500 250 338.29 856.99 241.01 1213.40 338.29 856.70 241.01 1212.90 338.29 856.28 241.01 1203.50

2.30 1000 1000 750 500 333.87 1688.10 236.58 1767.50 333.87 1687.70 236.58 1766.00 333.87 1686.90 234.37 1751.60

3.00 750 750 150 150 320.60 476.33 143.72 1014.80 320.60 476.07 141.51 999.55 320.60 475.71 141.51 999.51

VVdcdc and and PPdcdc outputs of 20x3 array configuration under partially shaded conditionsoutputs of 20x3 array configuration under partially shaded conditionsAssumed: The initial point for P&O is the voltage close to VAssumed: The initial point for P&O is the voltage close to Vococ

ANNANN--fuzzy logic with polar information based MPPT controlfuzzy logic with polar information based MPPT control

TestingTestingVerificationVerificationA. Under Uniform Insolation Condition

Different PV modules technologies:Different PV modules technologies:Siemens SMSiemens SM--55 (c55 (c--Si)Si)First Solar FSFirst Solar FS--50 (Thin50 (Thin--film CdTe)film CdTe)USSC USUSSC US--21 (3j a21 (3j a--Si)Si)

B. Under Partially Shaded Condition (SM-55)

Different PV array configurations:Different PV array configurations:SeriesSeries--Parallel (SP)Parallel (SP)Bridge Link (BL)Bridge Link (BL)Total Cross Tied (TCT)Total Cross Tied (TCT)Different PV array size:Different PV array size:3x3 (0.5kW)3x3 (0.5kW)20x3 (3.3kW)20x3 (3.3kW)

Developed real-time simulator based dSPACE real-time interface system

Overall results:Overall results:

Robust & insensitive to fast & Robust & insensitive to fast & slow variations in irradiance & slow variations in irradiance & temperature temperature It achieves to enhance the It achieves to enhance the dynamic behavior of the MPPT dynamic behavior of the MPPT controller without retuning the gain controller without retuning the gain parametersparameters

CONCLUSION

CONCLUSIONCONCLUSION

Performance of TFFN, RBF and ANFIS networks for different technology of PV modules

MarketsMarkets ofof nonnon--crystallinecrystalline siliconsilicon cellscellsConcentrationConcentration onon cc--SiSisolarsolar cellcell technologytechnologyDifferentDifferent MPPMPP pointspointscharacteristiccharacteristicEfficiencyEfficiency

MotivationMotivation--22

ANN structures: ANN structures: RBF, ANFIS, TFFNRBF, ANFIS, TFFN

Syafaruddin, Engin Karatepe, Takashi Hiyama: ‘Feasibility of Artificial Neural Network for Maximum Power Point Estimation of Non Crystalline-Si Photovoltaic Modules’, Proc. of The 15th International Conference on the Intelligent System Applications to Power Systems (ISAP), 8-11 November 2009, Curitiba, Brazil.

IntroductionIntroduction

• Investigation on three different ANN structures for identification the optimum operating voltage of non c-Si PV modules

• Type of PV modules: double junction amorphous Si (2j a-Si), triple junction amorphous Si (3j a-Si), Cadmium Indium Diselenide (CIS) and thin film Cadmium Telluride (CdTe) solar cell technologies

• Indicators for ANN models: the flexibility of training process, the simplicity of network structure and the accuracy of validation error

Description of the modelDescription of the model(Modeling of PV modules)(Modeling of PV modules)

• Expansion for non c-Si solar cells due to the development of nano-material technology

• The main reason of this trend is to cut the manufacturing cost of conventional Si PV modules

Other properties:Other properties:much cheaper for the same efficiency conversion, especially under massive production of cellslow costlight weightflexible more versatile with very small amount of Silicon neededenergy payback is very fast

(2 months, compared with 4 years for conventional Si technology)“In the end, the non c-Si solar cell technologies are potential to be inexpensive to produce and they could dominate the world production in the future”


PV module types: PV module types: Solarex MSTSolarex MST--43MV (2j a43MV (2j a--Si)Si)USSC UniSolar USUSSC UniSolar US--32 (3j a32 (3j a--Si)Si)Siemens STSiemens ST--40 (CIS)40 (CIS)First Solar FSFirst Solar FS--50 (thin film CdTe) 50 (thin film CdTe)

PVmodules

ISC(A)

VOC(V)

IMPP(A)

VMPP(V)

PMPP(W)

MST-43MVUS-32ST-40FS-50

0.7872.4

2.591.0

101.023.822.290

0.6161.942.410.77

7116.516.665

43.7432.0140.0050.05

Specifications under 1000W/mSpecifications under 1000W/m22 and 25and 25ooC C

Solarex MSTSolarex MST--43MV, USSC UniSolar US43MV, USSC UniSolar US--3232

(tandem junction and triple junction thin-film module amorphous silicon cells)“Effectively using the conversion of

sunlight”The bottom cell (red light), the middlecell (green light) and the top cell (bluelight)The major development: efficiency andstabilityOptimum conversion of differentsegments of the spectrum

Siemens ST-40(Cadmium Indium Diselenide (CIS) solar cells). characterized by exceptional spectral responseand long-term performance integrityEfficiency is almost similar to crystallinephotovoltaic modules

First Solar FSFirst Solar FS--5050(CdTe based thin-films technology)

•Recommended for high output voltage•very thin layers of compound semiconductor material with low temperature coefficients which provides for cost effective and greater energy production


The characteristics of IThe characteristics of Iscscand Vand Vococ are almost similar for are almost similar for all semiconductor types of all semiconductor types of solar cellssolar cellsHowever, there might be However, there might be different characteristics at different characteristics at VVMPPMPPThe correlation VThe correlation VMPPMPP=f(E,T=f(E,Tcc) ) is nonis non--linearlinear

“In this respect, to operate the “In this respect, to operate the PV modules at their maximum PV modules at their maximum power point, tracking the power point, tracking the optimum voltage using optimum voltage using intelligent technique by intelligent technique by means a functional reasoning means a functional reasoning method for function method for function approximation is totally approximation is totally necessary”necessary”

II--V curve characteristic model V curve characteristic model developed by Sandia National developed by Sandia National Laboratory Laboratory

ANN structures based optimum ANN structures based optimum voltagevoltage

Radial Basis Function (RBF):Radial Basis Function (RBF):••Strong networkStrong network••Fast training processFast training process••Direct confirmation of structureDirect confirmation of structure••But, end up with complicated structure, in But, end up with complicated structure, in case of complex datacase of complex data

Algorithm:Algorithm:•Local mapping, instead of global mapping as in MLP structure•In MLP; all inputs cause output; RBF; only input near the receptive fields produce the activation function (hidden layer)

))]1,())2,()1,(([)( 1111 nbTnwEnwdistradbasna c )]1,1())(),([()( 21 122bnxanmwpurelinma

n

n

The radial basis output layer a1: the Euclidean distance weight function ‘dist’ is applied between the input signals, [E;Tc] and weights, w1 before preceding them to the ‘radbas’ transfer function

The output layer aThe output layer a22: by simply applying : by simply applying the “purelin” transfer function between athe “purelin” transfer function between a11outputs and weights woutputs and weights w22, include the bias , include the bias bb22 of the second layer of the second layer

ANN structures based optimum voltage)ANN structures based optimum voltage)

Adaptive NeuroAdaptive Neuro--Fuzzy Inference System (ANFIS)Fuzzy Inference System (ANFIS)••Especially designed for single output, called Sugeno type fuzzy inference systemEspecially designed for single output, called Sugeno type fuzzy inference system••Hybrid learning algorithm (combine the least square and back propagation gradient descent Hybrid learning algorithm (combine the least square and back propagation gradient descent methods for training the mf parameters)methods for training the mf parameters)••Fast training process & high accuracyFast training process & high accuracy••But, for multiBut, for multi--objective optimization objective optimization multi anfis network and must be individually trained.multi anfis network and must be individually trained.

ANFIS network structures (cont.)ANFIS network structures (cont.)

LayerLayer--1:1:Grade of mfGrade of mf

)(1 xO iA

LayerLayer--2:2:Firing strengthFiring strength

m

jiAi xwO

12 )(

LayerLayer--3:3:Normalization of Normalization of

firing strengthfiring strength

213 ww

wwO ii

LayerLayer--4:4:Rule outputsRule outputs

)( 21114 iii rxqxpwyO

p,q,r are the coefficient parameters of the nth rule through the first order polynomial form expressed

in the first order Sugeno fuzzy model

LayerLayer--5:5:Single fixed nodeSingle fixed node

....)(

)(

222122

1211115

rxqxpw

rxqxpwyOii

i

)(11)( kIi ie

kO

j

jiji kOkwkI )()()(

2

1)()(

N

kkOktSSE

In the hidden and output layers, the In the hidden and output layers, the sigmoid function is utilized for the inputsigmoid function is utilized for the input--output characteristics of the nodes. For output characteristics of the nodes. For each node each node ii ; the output ; the output OOii(k)(k)

IIii(k)(k) : the input signal to node : the input signal to node ii at the at the kk--th th sampling and this is given by the weighted sampling and this is given by the weighted sum of the input nodessum of the input nodes

wwijij : the connection weight from node : the connection weight from node jj to node to node iiOOjj(k):(k): the output from nodethe output from node jj

During the training, the connection weights During the training, the connection weights wwijij are tuned recursively until the best fit is are tuned recursively until the best fit is achieved for the inputachieved for the input––output patterns output patterns based on the minimum value of the sum of based on the minimum value of the sum of the squared errors (SSE) the squared errors (SSE)

NN is the total number of training patterns, is the total number of training patterns, t(k)t(k) and and O(k)O(k) are the are the kk--th output target and estimated th output target and estimated values of MPPs voltagevalues of MPPs voltage--power power

During the training process:During the training process:learning rate = 0.2learning rate = 0.2

momentum rate = 0.85momentum rate = 0.85

ANN structures based ANN structures based optimum voltageoptimum voltage

Three layered FeedThree layered Feed--Forward Forward Network (TFFN):Network (TFFN):•Back propagation algorithm and the weight is adjusted by descent gradient method•Simple network with high accuracy•But, training process is time consuming with too many possible structures intuitive decision to select the best one


••To set the functions between input and To set the functions between input and output through the hidden layers output through the hidden layers (training (training data set should be selected to cover the data set should be selected to cover the entire region where the network is expected entire region where the network is expected to operate) to operate) ••Training data pattern Training data pattern [V[Vmppmpp]=f(E,T]=f(E,Tcc); ); following the mathematical model of SNLfollowing the mathematical model of SNL••Total data (228 sets) for operating conditions Total data (228 sets) for operating conditions between 15between 15--6565ooC and 100C and 100--1000W/m1000W/m22

PV modulesRBF ANFIS TFFN

nh SSE nh SSE nh SSE

MST-43MV 12 0.000375 21 0.004373 4 0.001429

US-32 4 0.000608 21 0.000146 4 0.000114

ST-40 5 0.000948 21 0.000914 4 0.000227

FS-50 11 0.000739 21 0.003154 7 0.001087

a. Ramp signal;E=90t+100 (W/m2) and Tc=5.5t+10 (oC); for 0 < t < 10sec

b. Random signal; where the mean of E and Tc are 800W/m2 and 50oC, respectively, with variance=1.0

c. Repeating sequence;E=450t+100 (W/m2) and Tc=27.5t+15 (oC); for the periodic time 0 < t < 2sec within 10sec of time simulation

d. Uniform random number; [Emin=100W/m2 -Emax=1000W/m2] and [Tcmin=10oC - Tcmax=65oC]

N

iMPP

iMPP

N

i

iMPP

iop VVVVJ

2

1

2 )()(

N is the number of parameter data set, i is the ith sample of data, VMPP is the ideal voltage at maximum power point, Vop is the estimated optimum voltage and

.

Typical results with random input signalsTypical results with random input signals

Solarex MST-43MV (2j a-Si)with RBF network

USSC Unisolar US-32 (3j a-Si)with ANFIS network

Siemens ST-40 (CIS)with TFFN network

First Solar FS-50 (thin film CdTe)with RBF network

SIMULATION RESULTS (cont.)

Performance index during validation processPerformance index during validation process

PV modulesramp signal random signal repeating sequence signal uniform random signal

RBF ANFIS TFFN RBF ANFIS TFFN RBF ANFIS TFFN RBF ANFIS TFFN

MST-43MV 0.035 0.096 0.065 0.258 0.982 0.992 0.035 0.113 0.070 0.035 0.092 0.065

US-32 0.143 0.007 0.059 2.474 0.036 0.418 0.128 0.007 0.059 0.146 0.008 0.060

ST-40 0.221 0.146 0.143 1.912 1.003 1.023 0.236 0.171 0.158 0.217 0.137 0.146

FS-50 0.063 0.115 0.081 0.205 2.009 0.868 0.071 0.133 0.064 0.067 0.112 0.080

Evaluation performance of ANN ModelsEvaluation performance of ANN Models

ANN models

Flexibility Training process

Simplicity of network structure

Accuracy of validation error

MST-43MV US-32 ST-40 FS-50

RBF high moderate high low moderate high

ANFIS high low low high high moderate

TFFN moderate high moderate moderate high high

ConclusionConclusion

Different models of artificial neural network to deal with the VDifferent models of artificial neural network to deal with the VMPPMPPof non crystalline Si PV modulesof non crystalline Si PV modules

The trained configurations are verified using ramp, random, The trained configurations are verified using ramp, random, repeating sequence and uniform random signals of irradiance and repeating sequence and uniform random signals of irradiance and cell temperaturecell temperature

The simulation results confirm that RBF and ANFIS methods have The simulation results confirm that RBF and ANFIS methods have the flexible training process; while the TFFN method has simpler the flexible training process; while the TFFN method has simpler network structure than others network structure than others

For the accuracy of validation error, RBF and ANFIS models are For the accuracy of validation error, RBF and ANFIS models are more suitable for 2j amore suitable for 2j a--Si and 3j aSi and 3j a--Si PV models, respectively. On Si PV models, respectively. On the other hand, ANFIS and TFFN are effectively used in CIS the other hand, ANFIS and TFFN are effectively used in CIS technology. For thin film technology. For thin film CdTeCdTe technology, the RBF and TFFN technology, the RBF and TFFN methods are the best optionmethods are the best option

Documents

ANN Methods in photovoltaic system MPPT controller ... · Maximum Power Point Tracking (MPPT) Control system. MPPT controller for uniform condition MPPT controller works based on