Design and Real Time Simulation of Artificial

Proceedings of the ASME 2014 International Mechanical Engineering Congress and Exposition IMECE2014 November 14-20, 2014, Montreal, Quebec, Canada

IMECE2014-37967 Design and Real Time Simulation of Artificial Intelligent based MPP Tracker for Photo-Voltaic System Shimi Sudha Letha Assistant Professor, NITTTR, Chandigarh, India Jagdish Kumar Associate Professor, PEC University of Technology, Chandigarh, India

Tilak Thakur Associate Professor, PEC University of Technology, Chandigarh, India Dnyaneshwar Karanjkar Research Scholar, NITTTR, Chandigarh, India

Abstract This paper presents an Artificial Intelligent based Maximum Power Point Tracking (MPPT) of a photo-voltaic system implementation using dSPACE 1104. The paper also proposes a novel Adaptive Neuro-Fuzzy Inference System (ANFIS) / Constant Voltage Tracker (CVT) for a photovoltaic (PV) powered multilevel inverter which requires a fixed constant dc voltage at its input. The MPPT algorithms viz. perturb and observe, incremental conductance, neural network, ANFIS and ANFIS/CVT have been designed and implemented on laboratory prototype. The modeling of various MPPT algorithms have been done on MATLAB/SIMULINK platform. Real time simulations have been carried out using dSPACE R&D controller board and CONTROLDESK software. The performance comparisons of various MPPT techniques applied to stand-alone PV system with resistive load have been presented for varying solar radiation conditions. The authors hope that the comparative analysis presented in this work will be helpful for further research.

Santanu Chatterji Professor and Head, NITTTR, Chandigarh,India

from renewable resources, the conventional biomass contributes to 10% and 3.4% from hydroelectricity. The small hydro, modern biomass, wind, solar, geothermal and biofuel account for another 3% and are expanding rapidly. The PV capacity worldwide at the end of 2012 was 100,000 MW. Germany is the world's leading solar PV installer with a capacity of 35.996 GW as of February 2014. The installed capacity of power in India at the end of February 2014 was 237.742 GW .The 87.55% of the installed capacity is constituted by the non renewable power plants, and the remaining 12.45% of total installed capacity is from the renewable power plants . India is endowed with an enormous solar energy potential. India receives about 5,000 trillion kWh of solar radiation per year with most parts receiving 4-7 kWh per m2 per day. Solar PV generation is a popular renewable energy source because of its unique features like absence of moving parts thus minimal maintenance, noise less, absence of fuel cost, emission free, reliable, clean and inexhaustible. It’s a known fact that the efficiency of the solar PV module is very low around 13%. Thus it is desirable to operate the PV module at its maximum power point so as to deliver maximum power to the load under varying irradiance and temperature levels. The solar PV module exhibit a nonlinear VI characteristics. It is essential to determine the duty cycle of the switch mode converter and control the voltage MPP or current MPP to automatically track the maximum power output using search algorithms at a given irradiance and temperature. Literature review reveal that, many MPP tracking (MPPT)

I. INTRODUCTION The energy scarcity and power quality are the major concern of this century thus most of the literature discuss the problems of natural resource depletion, environment impact, the rising demand of new energy resources and challenging technologies to overcome these problems. The projection by International Energy Agency (IEA) in 2011 shows that the solar power generators possibly will produce the world’s electricity within 50 years, reducing the emissions of greenhouse gases which harm the environment. Around 16% of global final energy consumption currently comes

1

Copyright © 2014 by ASME

Downloaded From: https://proceedings.asmedigitalcollection.asme.org/ on 10/10/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

methods like hill climbing / P&O (Perturb and Observe) [1-8], incremental conductance [9-12], current sweep are some of the fundamental approaches. Many other techniques like constant voltage [13], three points prediction algorithm [14], linear reoriented coordinates method (LRCM) [15], fractional-order incremental conductance method (FOICM) [16] , open-circuit voltage and short-circuit current [17], dual module V and I method [18] , ON/OFF control method [19], sensor less method [20], low cost method [21] , ripple correlation control, dc link capacitor droop control, Load voltage and load current maximization, dP/dV or dP/dI feedback control, β tracking, look up table method, one cycle control, variable inductor MPPT method, temperature gradient algorithm, smart optimal control using GA [22], biological swarm searching algorithm [23], artificial Intelligent techniques algorithms like fuzzy logic [24,25], neural network [26-29] and Adaptive Neuro Fuzzy Inference System (ANFIS) [30,31] etc. have been reported. The methods vary in sensed parameters, convergence speed, complexity, analog/digital control, PV array dependency, tuning requirement, cost, range of effectiveness, implementation complexity, popularity, and in other respects.

I and V are the PV array output current and voltage Iph is the photo-current that is equal to short circuit current I0 is the reverse saturation current np is the number of modules connected in parallel ns is the number of modules connected in series q is the charge of an electron k is Boltzmann’s constant a is the p-n junction ideality factor, 1< a < 5 (a = 1 being the ideal value), and T is the cell temperature. Rs

I +

Rsh

V

_ Fig. 1 Single Diode Model of a PV Cell The following solar module specifications from 'Vikram Solar' data sheet were used for simulating the PV model in MATLAB/SIMULINK shown in Fig. 2.

The MATLAB/SIMULINK model of perturb and observe, incremental conductance, neural network, ANFIS and the proposed ANFIS / CVT for a PV powered multilevel inverter have been developed and are experimentally validated. The proposed MPPT scheme is simple and tracks a constant 12V dc voltage for voltage above 12 volts and tracks an ANFIS based maximum power point for voltage less than 12 V. The rest of the paper is organized as follows. The section II presents the modeling and characteristics of PV module. In section III the MPPT algorithms Perturb and Observe, Incremental Conductance, Neural Network based MPPT and ANFIS (Adaptive Neuro Fuzzy Inference System) based MPPT for a PV module are explained, followed by the proposed ANFIS / CVT in section IV. The maximum power prediction model is given in section V. To validate the simulation result the hardware prototype is implemented, and the experimental results are presented in Section VI. Finally, the paper is concluded in Section VII.

TABLE 1: VIKRAM SOLAR MODULES SPECIFICATIONS

Description Number of cells in series (nCells) Maximum power ( Pmp) Maximum power voltage (Vmp) Maximum power current (Imp) Open circuit voltage (Voc) Short circuit current (Isc) Maximum power temp. coefficient (TempC_Pmp) Maximum power voltage temp. coefficient (TempC_Vmp) Maximum power current temp. coefficient (TempC_Imp) Open circuit voltage temp. coefficient (TempC_Voc) Short circuit current temp. coefficient (TempC_Isc) Series resistance of PV model (Rs) Parallel resistance of PV model (Rp) Diode saturation current of PV model (Isat) Light-generated photo-current of PV model ( Iph) Diode quality factor of PV model Qd

II MODELING AND CHARACTERISTICS OF PV MODULE Modeling of a solar cell is done by connecting a current source in parallel with an inverted diode along with a series and a parallel resistance as shown in Fig.1. The series resistance is due to hindrance in the path of flow of electrons from n to p junction and parallel resistance is due to the leakage current. It has non-linear characteristics, and the basic mathematical model can be expressed as: (1)

Vikram Solar 36 37 W 17 V 2.25A 21.77 V 2.26 A -9.770e-001 W/deg.C -1.230e-001 V/deg.C -1.079e-003 A/deg.C -1.200e-001 V/deg.C -5.016e-003 A/deg.C 0.450 ohms 1442.84 ohms 2.6762e-006 A 2.4 A 1.5

where,

2

Copyright © 2014 by ASME

Downloaded From: https://proceedings.asmedigitalcollection.asme.org/ on 10/10/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

Fig. 4 PV Characteristics of the PV Module

Fig.2 MATLAB/SIMULINK Model of PV Module The diode characteristics is simulated using the following equation. Id=Isat*exp(Vd/VT -1) (2) Where, Id = Diode current (A) Vd = Diode voltage (V) Isat = Diode saturation current (A) VT = Temperature voltage = k*T/q*Qd*Ncell*Nser T = Cell temperature (K), k = Boltzman constant = 1.3806e-23 J.K^-1 q = Electron charge = 1.6022e-19 C Qd = Diode quality factor Ncell= Number of series-connected cells per module Nser = Number of series-connected modules per string

Fig. 5 VI Characteristics of the PV Module

The MATLAB/SIMULINK model of the diode characteristics is shown in Fig.3 .

III. MPPT ALGORITHMS The electrical power generated by a PV system depends on solar irradiance , temperature and cloud cover. The current and voltage at which a solar module generates the maximum power is known as the maximum power point which is not known in advance. The maximum power point tracking (MPPT) system basically controls the voltage and the current of the PV array independently of the connected load. MPPT algorithms Perturb and Observe, Incremental Conductance, Neural Network and Adaptive Neuro Fuzzy Inference System (ANFIS) which are implemented on the PV array are described below [32]. 1.Perturb and Observe In this method a slight perturbation is introduced in the system. Thus, the power of the module alters. Due to the perturbation if the power enhances, then the perturbation is carried on in the same direction. After the maximum power is accomplished, the power at the next instant decrements and hence the perturbation reverses. As in this method only one voltage and current sensor is used the error due to change in irradiance is introduced.

Fig. 3 MATLAB/SIMULINK Model of the Diode Characteristics The PV and VI characteristics of the PV module for varying irradiations (1KW/m2, 0.75KW/m2, 0.5KW/m2, 0.25KW/m2) and constant temperature 25oC are given in Fig. 4 and Fig. 5 respectively.

3

Copyright © 2014 by ASME

Downloaded From: https://proceedings.asmedigitalcollection.asme.org/ on 10/10/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

Using the collected data as input of artificial neural network (ANN) and the obtained emulated MPP locus (EML) parameter as output of artificial neural network (ANN), the proposed artificial neural network (ANN) can be trained using the Neural Network Toolbox in MATLAB [34].

2. Incremental Conductance Method Incremental conductance method senses both the voltage and current of the PV array simultaneously thus the error due to change in irradiance is eradicated. At MPP the slope of the PV curve is 0. (dP/dV)MPP= d(VI)/dV 0= I+VdI/dVMPP dI/dVMPP = - I/V (3) The left hand side of equation 3 is the instantaneous conductance of the solar panel. When this instantaneous conductance equals the conductance of the PV array then MPP is reached. This method is more complex and the cost of implementation is also high and this method is suitable for a highly elaborated system. 3. Neural Network based MPPT Neural network architecture gives the appropriate solution for the non-linear and complex systems. The simpler architecture contains two layers as shown in Fig. 6.

A two-layer feed-forward network with sigmoid hidden neurons and linear output neurons, can fit multi-dimensional mapping problems arbitrarily well, given consistent data and enough neurons in its hidden layer. The network was trained with Levenberg-Marquardt backpropagation algorithm. A total of 7609 samples were collected from real time system out of which 5327 samples (ie) 70% were used as training data, 1141 samples (ie) 15% were used as validation data and remaining 1141 samples (ie) 15% were used as testing data. The mean square error (MSE) and regression of the ANN model developed is given in Table 2. TABLE 2: ANN DATA SET, MEAN SQUARE ERROR AND REGRESSION

Fig 6. Neural Network Model In this paper, to map the relationship between the emulated MPP locus (EML) parameter sets and the parameters available from the datasheet, a two layer feed-forward ANN as shown in Fig. 6 is utilized. The input layer in this case consists of PV voltage and current recorded under different irradiations and temperatures from a real time system. The number of hidden layer nodes is 10. The relationships between the inputs and outputs are given as follows [33]:

Samples

MSE

Regression

Training

5327

2.92492e-9

1.90887e-1

Validation

1141

2.82180e-9

1.93351e-1

Testing

1141

2.88905e-9

1.79767e-1

To validate the ANN model the error histogram is shown in Fig. 7. Error Histogram with 20 Bins 3500

Training Validation Test Zero Error

3000

Instances

2500

Neurons in input layer only act as buffers for distributing the input signals. The net input of the jth hidden unit is (4) where is the weighting on the connection from the ith input unit, represent the bias for the hidden layer neurons. The output of the neurons in the hidden layer is (5) and the net input to the neurons in the output layer can be written as (6) where is the weighting on the connection from the jth represent the bias for the second layer neurons. input unit, , is the network outputs of The output of the second layer, interest, and these outputs are labeled yk . x (7) The collected data is first processed by MATLAB to obtain the corresponding emulated MPP locus (EML) parameter set.

2000 1500 1000

7.2e-05

6.28e-05

5.37e-05

4.46e-05

3.55e-05

2.64e-05

1.73e-05

8.16e-06

-1e-05

-9.5e-07

-1.9e-05

-2.8e-05

-3.7e-05

-4.7e-05

-5.6e-05

-6.5e-05

-7.4e-05

-8.3e-05

-0.0001

0

-9.2e-05

500

Errors = Targets - Outputs

Fig. 7. Validation of the ANN model ( Error Histogram) The two-layer feed-forward network with sigmoid hidden neurons and linear output neurons based artificial neural network (ANN) model developed using MATLAB/SIMULINK is shown in Fig.8.

4

Copyright © 2014 by ASME

Downloaded From: https://proceedings.asmedigitalcollection.asme.org/ on 10/10/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

The mapping of training data and ANFIS output , membership function of ANFIS inputs current and membership function of ANFIS inputs voltage are shown in Fig. 10, Fig. 11 and Fig. 12 respectively.

Fig. 8. MATLAB/SIMULINK ANN Model 4. ANFIS (Adaptive Neuro Fuzzy Inference System) Intelligent control is the viable alternative to conventional control schemes. The uncertain or unknown variations in plant parameters can be dealt more effectively by using artificial intelligent techniques such as fuzzy logic and neural network. Hence the robustness of the control system can be improved. The multilayer feed forward network has nodes which performs a particular function on incoming signals. Each node has different formula. The links in the adaptive network just indicate the signal flow direction [35-37]. The Investigators have developed an Adaptive Neuro Fuzzy Inference System (ANFIS) based algorithm for Maximum Power Point Tracking (MPPT). Voltage , current and the duty cycle was recorded under different irradiations and temperatures from a real time system. This real time data was used to train the Sugeno-type Adaptive Neuro Fuzzy Inference System (ANFIS) model using MATLAB/SIMULINK. A hybrid learning algorithm which is a combination of the least-squares method and the backpropagation gradient descent method is used for training FIS membership function parameters to emulate a given training data set. Different membership function such as triangular, trapizodal, gaussian etc. were tried out of which the gaussian membership function for both the inputs gave minimum training error of 8.276e-6 in 20 epochs. The membership function type for output data was choosen as constant.

Fig. 10 Training Data and ANFIS Output

Fig. 11 Membership Function of ANFIS Input (Current)

Fig. 12 Membership Function of ANFIS Input (Voltage) The nine fuzzy inference rules that shows the input and output relations are as follows. (i) If (current is in1mf1) and (voltage is in 2mf1) then (duty cycle is out1mf1) (ii) If (current is in1mf1) and (voltage is in2mf2) then (duty cycle is out1mf2) (iii) If (current is in1mf1) and (voltage is in2mf3) then (duty cycle is out1mf3) (iv) If (current is in1mf2) and (voltage is in2mf1) then (duty cycle is out1mf4) (v) If (current is in1mf2) and (voltage is in2mf2) then (duty cycle is out1mf5)

Fig. 9. MATLAB/SIMULINK ANFIS Model

5

Copyright © 2014 by ASME

Downloaded From: https://proceedings.asmedigitalcollection.asme.org/ on 10/10/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

The voltage and current sensors are used to acquire the input PV voltage and current and is fed as input to the real time interface unit ds1104. Dspace 1104 is interfaced with MATLAB/SIMULINK in PC, which generates the firing pulse of required duty cycle to fire the buck converter. For the proposed MPPT algorithm to track a constant 12 V dc voltage for a PV powered multilevel inverter the output voltage is sensed using the output voltage sensor and compared with a 12 V reference, based on the error signal a duty cycle is generated. The radiation and temperature sensors are used to calculate the maximum power output of the PV panel to calculate the PV module efficiency.

(vi) If (current is in1mf2) and (voltage is in2mf3) then (duty cycle is out1mf6) (vii) If (current is in1mf3) and (voltage is in2mf1) then (duty cycle is out1mf7) (viii) If (current is in1mf3) and (voltage is in2mf2) then (duty cycle is out1mf8) (ix) If (current is in1mf3) and (voltage is in2mf3) then (duty cycle is out1mf9) IV. PROPOSED ANFIS / CONSTANT VOLTAGE TRACKER (CVT) ALGORITHM A solar powered cascade multilevel inverter require a constant and isolated dc power supply. This dc supply is obtained using a PV array with nonlinear characteristics. Thus the proposed algorithm can maintain any required constant voltage depending upon the application. As the authors have planned to further extend this work to make a prototype of cascade multilevel inverter with a constant and isolated 12 V PV input, a 12V dc input tracking has been discussed further. The duty cycle of the buck converter is determined using the following equation when voltage is equal to or more than 12V, d = K (Vout - Vref) (8) where, d is the duty cycle, Vout is the output voltage of the PV array, Vref is the reference voltage and K is the gain of the amplifier. For voltage less than 12V the module tracks the MPP using ANFIS algorithm. The most popularly used MPPT algorithms Perturb and Observe ,Incremental Conductance, Neural Network based MPPT , ANFIS based MPPT algorithms and the proposed MPPT algorithm to obtain a constant 12 V dc voltage for a PV powered multilevel inverter are developed for generating the duty cycle of the buck converter. The block diagram of the proposed scheme is shown in Fig.13.

The MATLAB/SIMULINK model of the proposed model is shown in the Fig.14. For voltage more than 12V the algorithm tracks 12V and for voltage less than 12V the algorithm tracks the maximum power point according to ANFIS algorithm.

PV Current Sensor

The output voltage and output power corresponding to the changing irradiance are shown in Fig.15 and Fig. 16 respectively.

Load

Radiation Sensor

Buck Converter

Fig.14 MATLAB/SIMULINK Model of Constant Voltage MPPT Tracker

PV Array

Duty Cycle

Temperature Sensor

PV Voltage Sensor

Dspace ds1104

PC with MATLAB/ SIMULINK (MPPT Algorithm)

Output Voltage Sensor

Fig.15 Output Voltage Corresponding to Irradiance Level Fig. 13. Block Diagram of the Proposed System

6

Copyright © 2014 by ASME

Downloaded From: https://proceedings.asmedigitalcollection.asme.org/ on 10/10/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

circuit voltage and fill factor are shown in Fig.17, Fig.18 and Fig.19 respectively.

Fig.16 Output Power Corresponding to Irradiance Level Fig.18 Sub-system for Open Circuit Voltage

V. Maximum Power Prediction Model The maximum power output Pmax of a PV module can be expressed in terms of three important electrical characteristics (i) Short Circuit Current Isc (ii) Open Circuit Voltage Voc and Fill factor FF as functions of the irradiation and module temperature [38]. The maximum power output Pmax o f a PV module can be written as in equation (9) and (10), (9)

Fig.19 Sub-system for Short Circuit Current The response of maximum power, open circuit voltage, short circuit current and fill factor for the change in irradiance level are as shown in Fig 20.

(10) Where, Isco = Short-circuit current under standard solar irradiance Go Isc = Short-circuit current under solar irradiance G , & = Constant parameters for PV module = Open-circuit voltage under normal solar irradiance = Open-circuit voltage under Go FF0 = Fill factor without resistive effects = Series resistance = Normalized value of open-circuit voltage n = Ideality factor (1