DRL-based Controller Design for Buck-Boost Converter

This project explores the use of Deep Reinforcement Learning (DRL) to design and evaluate controllers for a Buck-Boost Converter. The Buck-Boost Converter is a DC-DC converter used in various applications like load converters for electronic devices, electric vehicles, and photovoltaic cells.

Overview

• Objective: To control a Buck-Boost Converter efficiently using DRL algorithms and compare their performance.
• Algorithms Implemented:
  • Actor-Critic (AC)
  • Proximal Policy Optimization (PPO)
  • Deep Deterministic Policy Gradient (DDPG)
  • Hybrid model combining DRL with PID control (AC and DDPG)

Key Concepts

• Deep Reinforcement Learning (DRL): Applied to optimize the control strategy of the Buck-Boost Converter.
• Actor-Critic Architecture: Used to improve the policy based on real-time feedback.
• PPO & DDPG: Advanced RL algorithms tailored for continuous action spaces in power electronic systems.

Steps to Recreate the Results

1. Setup the Environment:

   • Ensure you have MATLAB and Simulink installed (R2023a or later recommended).
   • Clone or download the repository containing the MATLAB scripts and Simulink models.

2. Folder Structure:

   • Place all MATLAB files (.m files) and Simulink models (.slx files) in the same directory.

3. Running the Simulations:

   • Open MATLAB and navigate to the folder containing the project files.
   • Run the desired MATLAB script (e.g., run_drl_simulation.m). Ensure that the corresponding Simulink model is in the same folder so that it can be accessed during the simulation.
   • The scripts will automatically load the Simulink models and execute the simulations.

4. Analyzing Results:

   • Upon completion, the results will be displayed in MATLAB, showing the performance metrics of the different DRL algorithms.
   • You can modify the parameters in the scripts to test different scenarios and compare results.

Results

• Comparative analysis of different RL-based control strategies showed improved efficiency and stability over traditional methods.

Acknowledgments

This project was supervised by Dr. Sudha Radhika as part of the Power Electronics course at BITS Pilani, Hyderabad Campus.

License

This project is licensed under the MIT License.