DDPG_FDI_SG

Enhancing Smart Grid Cyber Resilience against FDI attacks using Multi-Agent Recurrent DDPG

Contributions: Dr. Tahira Mahboob (Netlab - UofG): tahira.mahboob@yahoo.com, Mingwei Li (UoE): ml969@exeter.ac.uk

PandaPower (Pandapower + RL)

This repository contains:

• Legacy training/evaluation scripts based on pandapower.timeseries.run_timeseries and custom controllers.
• A PettingZoo/RLlib path (multi-agent) for RL training.

Setup

Create and activate venv:

python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

Legacy scripts (run_timeseries-based)

Run DDPG training:

python tests/ddpg_training.py

Run LSTM-DDPG training:

python tests/ddpg_lstm/ddpg_lstm_training.py

Run DDPG evaluation (loads saved actor checkpoints):

python tests/ddpg_test.py

Run LSTM-DDPG evaluation:

python tests/ddpg_lstm/ddpg_lstm_test.py

PettingZoo env smoke test

python tests/rllib_smoke_test.py