PSO Trained Neural Network MPPT for Solar PV System

PSO Trained Neural Network MPPT for Solar PV System

𝐈𝐧𝐭𝐫𝐨𝐝𝐮𝐜𝐭𝐢𝐨𝐧

Solar photovoltaic output changes continuously with solar irradiance, temperature, and load conditions. Therefore, an effective Maximum Power Point Tracking technique is required to extract the highest available power from the PV panel.

PSO Trained Neural Network MPPT for Solar PV System

PSO Trained Neural Network MPPT for Solar PV system

₹5,000.00₹2,500.00

Buy Now

The PSO Trained Neural Network MPPT combines:

• Particle Swarm Optimization (PSO) for neural-network training

• Artificial Neural Network (ANN) for maximum-power-point voltage prediction

• PID control for duty-cycle generation

• Boost converter for PV voltage regulation

• PWM control for converter switching

The complete system is modeled and tested in MATLAB/Simulink under changing load and solar-irradiance conditions.

𝐒𝐲𝐬𝐭𝐞𝐦 𝐎𝐯𝐞𝐫𝐯𝐢𝐞𝐰

The proposed solar PV system contains the following major components:

• 250 W solar PV panel

• PSO-trained neural network

• PID voltage controller

• PWM generator

• DC–DC boost converter

• Variable resistive load

• Voltage, current, and power measurement blocks

System Configuration

Parameter	Specification
PV panel rated power	250 W
Open-circuit voltage	37.3 V
Voltage at maximum power point	30.7 V
Short-circuit current	8.66 A
Current at maximum power point	8.15 A
Standard irradiance	1000 W/m²
Standard temperature	25°C
Converter type	DC–DC boost converter
MPPT method	PSO-trained neural network
Controller	PID controller
Implementation platform	MATLAB/Simulink

𝐀𝐫𝐭𝐢𝐟𝐢𝐜𝐢𝐚𝐥 𝐍𝐞𝐮𝐫𝐚𝐥 𝐍𝐞𝐭𝐰𝐨𝐫𝐤 𝐒𝐭𝐫𝐮𝐜𝐭𝐮𝐫𝐞

The artificial neural network consists of three basic sections:

• Input layer

• Hidden layer

• Output layer

Each neuron is connected using adjustable weights and biases. These parameters determine how accurately the neural network can predict the required maximum-power-point voltage.

Neural Network Inputs and Output

ANN Variable	Signal
Input 1	Solar irradiance
Input 2	PV panel temperature
Target output	Voltage at maximum power point
Final ANN output	Reference PV voltage

The trained neural network predicts the optimum PV voltage for the measured irradiance and temperature.

𝐖𝐡𝐲 𝐏𝐒𝐎 𝐈𝐬 𝐔𝐬𝐞𝐝

Conventional neural networks commonly update their weights and biases using standard training algorithms. In this system, Particle Swarm Optimization is used to optimize these parameters.

PSO searches for a suitable combination of:

• Input-to-hidden-layer weights

• Hidden-to-output-layer weights

• Hidden-layer biases

• Output-layer bias

The objective is to minimize the difference between:

• Expected maximum-power-point voltage

• Neural-network-predicted voltage

PSO Training Settings

PSO Parameter	Value
Maximum iterations	100
Population size	100 particles
Optimized variables	ANN weights and biases
Performance measure	Root mean square error
Reported minimum cost	Approximately 0.0161

A cost value closer to zero indicates better agreement between the neural-network output and its target value.

𝐓𝐫𝐚𝐢𝐧𝐢𝐧𝐠 𝐃𝐚𝐭𝐚 𝐆𝐞𝐧𝐞𝐫𝐚𝐭𝐢𝐨𝐧

Training data are generated using a mathematical solar PV panel model in MATLAB.

The model produces operating data for different combinations of:

• Solar irradiance

• Cell temperature

• Maximum-power-point current

• Maximum-power-point voltage

• Maximum available PV power

Dataset Operating Range

Data Variable	Range
Solar irradiance	0–1000 W/m²
Temperature	15–35°C
ANN input data	Irradiance and temperature
ANN target data	Maximum-power-point voltage

Random irradiance and temperature values are generated within the specified ranges. The PV model then calculates the corresponding maximum-power-point voltage.

This procedure creates a diverse dataset for training the neural network under different environmental conditions.

𝐏𝐒𝐎-𝐁𝐚𝐬𝐞𝐝 𝐀𝐍𝐍 𝐓𝐫𝐚𝐢𝐧𝐢𝐧𝐠 𝐏𝐫𝐨𝐜𝐞𝐬𝐬

The PSO training procedure follows these main steps:

1. Generate PV operating data.

2. Arrange irradiance and temperature as ANN inputs.

3. Use maximum-power-point voltage as the target.

4. Configure the feedforward neural network.

5. Extract its initial weights and biases.

6. Represent the weights and biases as PSO particles.

7. Calculate the ANN output for every particle.

8. Compare the predicted voltage with the target voltage.

9. Calculate the training error.

10. Update particle velocities and positions.

11. Repeat the optimization until the maximum iteration is reached.

12. Assign the best weights and biases to the neural network.

13. Convert the trained network into a Simulink-compatible block.

The resulting neural network can predict the optimum PV operating voltage without performing a slow online search.

𝐖𝐨𝐫𝐤𝐢𝐧𝐠 𝐏𝐫𝐨𝐜𝐞𝐬𝐬

The operation of the complete PV system can be summarized as follows:

1. The irradiance and temperature are measured.

2. These signals are supplied to the PSO-trained ANN.

3. The ANN predicts the maximum-power-point voltage.

4. The predicted voltage becomes the reference voltage.

5. The reference voltage is compared with the measured PV voltage.

6. The voltage error is processed by the PID controller.

7. The controller produces the required duty-cycle command.

8. The PWM generator creates switching pulses.

9. The pulses control the boost-converter switch.

10. The PV panel is regulated near its maximum power point.

𝐂𝐨𝐧𝐭𝐫𝐨𝐥 𝐒𝐭𝐫𝐚𝐭𝐞𝐠𝐲

The control system contains three coordinated stages.

1. Maximum-Power-Point Prediction

The PSO-trained neural network estimates the optimum PV voltage from the environmental inputs.

2. Voltage Regulation

The PID controller compares:

• ANN-generated reference voltage

• Actual PV panel voltage

It processes the error and generates the appropriate converter duty cycle.

3. PWM Generation

The duty-cycle signal is supplied to the PWM generator. The resulting gate pulse controls the boost converter and maintains the required PV operating point.

Control Signal Summary

Control Stage	Input	Output
PSO-trained ANN	Irradiance and temperature	Reference PV voltage
Voltage comparator	Reference and measured PV voltage	Voltage error
PID controller	Voltage error	Duty cycle
PWM generator	Duty cycle	Switching pulse
Boost converter	PV input and switching pulse	Regulated DC output

𝐏𝐕 𝐏𝐚𝐧𝐞𝐥 𝐂𝐡𝐚𝐫𝐚𝐜𝐭𝐞𝐫𝐢𝐬𝐭𝐢𝐜𝐬

At a constant temperature of 25°C, the maximum available PV power changes with solar irradiance.

Maximum Power at Different Irradiance Levels

Irradiance	Approximate Maximum Power
1000 W/m²	250.0 W
800 W/m²	199.9 W
600 W/m²	149.6 W
400 W/m²	98.9 W
200 W/m²	48.37 W

These values provide useful reference points for evaluating the tracking performance of the proposed MPPT controller.

𝐒𝐢𝐦𝐮𝐥𝐚𝐭𝐢𝐨𝐧 𝐓𝐞𝐬𝐭 𝐂𝐨𝐧𝐝𝐢𝐭𝐢𝐨𝐧𝐬

The system is evaluated under two important operating conditions:

• Sudden load variation

• Changing solar irradiance

𝐓𝐞𝐬𝐭 𝟏: 𝐕𝐚𝐫𝐢𝐚𝐛𝐥𝐞 𝐋𝐨𝐚𝐝 𝐂𝐨𝐧𝐝𝐢𝐭𝐢𝐨𝐧

For this test, solar irradiance and temperature are maintained at standard values while the connected load is changed.

Variable-Load Test Settings

Parameter	Test Condition
Solar irradiance	1000 W/m²
Temperature	25°C
Load switching interval	0.3 seconds
Number of loads	Three
Expected PV maximum power	Approximately 250 W

Load-Switching Sequence

Time Period	Connected Load
Initial to 0.3 s	Load R1
0.3 to 0.6 s	Load R2
After 0.6 s	Load R3

The simulation monitors:

• PV voltage

• PV current

• PV power

• Load voltage

• Load current

• Load power

Observed Performance

• The PV power remains close to 250 W.

• The controller quickly responds when the load changes.

• PV voltage is maintained near the maximum-power-point voltage.

• PV current remains close to the required maximum-power-point current.

• Short transients appear during load switching.

• The system returns rapidly to the desired operating point.

This test demonstrates that the PSO-trained ANN MPPT can maintain maximum power extraction during sudden load changes.

𝐓𝐞𝐬𝐭 𝟐: 𝐕𝐚𝐫𝐢𝐚𝐛𝐞 𝐈𝐫𝐫𝐚𝐝𝐢𝐚𝐧𝐜𝐞 𝐂𝐨𝐧𝐝𝐢𝐭𝐢𝐨𝐧

For the second test, the load is maintained constant while the solar irradiance is changed every 0.2 seconds.

Irradiance Test Sequence

Time Interval	Irradiance	Expected Maximum Power
0–0.2 s	1000 W/m²	Approximately 250 W
0.2–0.4 s	800 W/m²	Approximately 200 W
0.4–0.6 s	600 W/m²	Approximately 150 W
0.6–0.8 s	400 W/m²	Approximately 100 W
0.8–1.0 s	200 W/m²	Approximately 50 W

Observed Tracking Results

Irradiance	PV Characteristic Value	Simulated Tracking Level
1000 W/m²	250.0 W	Around 250 W
800 W/m²	199.9 W	Around 200 W
600 W/m²	149.6 W	Around 150 W
400 W/m²	98.9 W	Around 100 W
200 W/m²	48.37 W	Around 48–50 W

The close agreement between the expected and simulated power levels confirms effective maximum power point tracking.

𝐒𝐢𝐦𝐮𝐥𝐚𝐭𝐢𝐨𝐧 𝐑𝐞𝐬𝐮𝐥𝐭𝐬

The MATLAB/Simulink results show that the PSO-trained neural network can successfully estimate the required PV reference voltage.

Main Performance Observations

• Maximum power is extracted under standard operating conditions.

• Tracking remains effective during sudden load variations.

• The ANN adapts its reference voltage when irradiance changes.

• PV power closely follows the panel’s theoretical maximum-power values.

• Only small transient variations occur during operating-point changes.

• The boost converter maintains the required power transfer to the load.

• The PID controller provides stable duty-cycle adjustment.

• PSO training reduces the ANN prediction error.

𝐏𝐒𝐎-𝐓𝐫𝐚𝐢𝐧𝐞𝐝 𝐀𝐍𝐍 𝐯𝐬. 𝐂𝐨𝐧𝐯𝐞𝐧𝐭𝐢𝐨𝐧𝐚𝐥 𝐌𝐏𝐏𝐓

Feature	Conventional MPPT	PSO-Trained ANN MPPT
Operating principle	Repeated online searching	Learned voltage prediction
Primary inputs	PV voltage and current	Irradiance and temperature
Reference generation	Perturbation-based	ANN-based
Offline optimization	Usually not required	PSO training required
Response to environmental changes	Depends on search speed	Rapid reference prediction
Steady-state oscillation	May be present	Can be reduced
Implementation complexity	Relatively simple	Moderately advanced
Suitable for intelligent control studies	Limited	Highly suitable

𝐊𝐞𝐲 𝐅𝐞𝐚𝐭𝐮𝐫𝐞𝐬

• PSO-optimized neural-network parameters

• Irradiance and temperature-based MPPT

• Prediction of maximum-power-point voltage

• MATLAB-based PV dataset generation

• Simulink-compatible trained ANN block

• PID-controlled boost converter

• PWM switching-pulse generation

• Variable-load performance analysis

• Step-change irradiance testing

• Voltage, current, and power monitoring

• Suitable for a 250 W solar PV panel

• Clear comparison with PV characteristic data

𝐀𝐝𝐯𝐚𝐧𝐭𝐚𝐠𝐞𝐬

• Provides fast maximum-power-point voltage estimation

• Reduces dependency on continuous perturbation

• Offers good performance under changing irradiance

• Maintains power extraction during sudden load changes

• Combines optimization and machine-learning techniques

• Can be integrated into larger renewable-energy systems

• Supports detailed controller-performance analysis

• Helps researchers study intelligent MPPT techniques

𝐀𝐩𝐩𝐥𝐢𝐜𝐚𝐭𝐢𝐨𝐧𝐬

The PSO-trained neural network MPPT method can be applied in:

• Standalone solar PV systems

• Solar battery-charging systems

• DC microgrids

• Solar-powered electric-vehicle charging

• Hybrid renewable-energy systems

• PV-fed DC motor drives

• Solar water-pumping systems

• Grid-connected PV converters

• Intelligent power-electronic controllers

• Renewable-energy research and laboratory studies

𝐋𝐞𝐚𝐫𝐧𝐢𝐧𝐠 𝐎𝐮𝐭𝐜𝐨𝐦𝐞𝐬

By studying this MATLAB/Simulink model, learners can understand:

• Solar PV panel modeling

• PV dataset generation

• ANN input and target preparation

• Feedforward neural-network configuration

• Optimization of ANN weights and biases using PSO

• Cost-function minimization

• ANN conversion into a Simulink block

• PID voltage regulation

• Boost-converter control

• PWM pulse generation

• MPPT testing under different operating conditions

• Interpretation of PV voltage, current, and power plots

𝐂𝐨𝐧𝐜𝐥𝐮𝐬𝐢𝐨𝐧

The PSO Trained Neural Network MPPT for Solar PV System provides an intelligent approach for extracting the maximum available power from a photovoltaic panel.

PSO optimizes the neural network’s weights and biases, while the trained ANN predicts the maximum-power-point voltage using solar irradiance and temperature. A PID-controlled boost converter then regulates the PV panel at the predicted operating point.

The simulation results demonstrate effective performance under both changing load and changing irradiance conditions. The extracted PV power closely follows the expected maximum-power values, confirming the effectiveness of the proposed MPPT method.

This MATLAB/Simulink implementation is useful for students, researchers, and engineers studying solar PV control, neural networks, optimization algorithms, DC–DC converters, and intelligent MPPT techniques.