Detection and Mitigation of Bad Data Injection Attack using Deep Reinforcement Learning in a Virtual Power Plant

With the growing complexity of energy systems, Virtual Power Plants (VPPs) have become essential for efficient renewable energy distribution. However, their dependence on interconnected resources exposes them to various cyber threats, such as Bad Data Injection (BDI) attacks. This work implements a BDI attack detection-mitigation strategy using Reinforcement Learning (RL) for detection, and an Adaptive Proportional-Integral (PI) optimized using Deep RL for mitigation. The BDI attack was simulated in Simulink on the Battery Energy Storage System of the VPP for the training dataset. This dataset was implemented in Python for various detection models, and a Deep Q-Network achieved scores of 97% for Accuracy, Precision, Recall, and F1-score. The best mitigation controller was found to be the Adaptive PI controller using a Deep RL agent using the Deep-Deterministic Policy-Gradient algorithm, resulting in best values of State-of-Charge, Total Harmonic Distortions, and Voltage Index values of 50.014%, 0.30%, and 1.21, respectively.