Deep Reinforcement Learning for Power Distribution System Reconfiguration

This repository implements three deep reinforcement learning (RL) algorithms: PPO, A2C, and TRPO, to reconfigure power distribution systems and maximize load supply during disasters. This work was part of a class project at NYU.

Overview

• Environment:

  • The IEEE 34 test system is used as the environment for the RL algorithms.
  • The environment is sourced from this repository with a few modifications to make it self sufficient.

• Modifications:

  • Added a Convolutional Neural Network (CNN) as a feature extractor.
  • Conducted comprehensive simulation cases in both normal and post-disaster conditions.
  • Tested A2C and TRPO algorithms alongside PPO.

• Report:

  • Please refer to the attached 'project report' for detailed analysis and results.

Instructions for Running the Code

• Notebooks:

  • Two Jupyter notebooks are provided for running the simulations:
    • run_normal.ipynb: Reconfigure the system during normal conditions.
    • run_post_fault.ipynb: Reconfigure the system during post-disaster conditions.

• Environment Setup:

  • The following files are related to the RL environment settings:
    • DGs.py, DSS_CircuitSetup.py, DSS_Initialize.py, ieee34Mod1.dss, IEEELineCodes.dss, openDSSenv34.py, state_action_reward.py, switch.py.

Additional Requirements

• TensorBoard:

  • Required to view the figures generated after the simulation.

• Operating System:

  • As the OpenDSS software is supported on Windows, this repository requires a Windows PC to run.

• Python Version:

  • The code is tested with Python 3.8.10.

• Dependencies:

  • Install the required Python libraries from the requirements.txt file.

  pip install -r requirements.txt
  

Contact

• Zahidul Islam (mi2502@nyu.edu)