Evaluation of Optimal Strategy under False Data Injection Attack On Power System: A Reinforcement Learning Approach

Abstract

Conventional power grids are transforming into more automated smart power grids, with a significant em-phasis on improved performance and enhanced reliability. However, due to the substantial incorporation of communication networks, this cyber-physical system (CPS) is vulnerable to frequent cyber threats. During the information interchange, the invader might penetrate the power system and introduce false data, causing intentional system instability, power outages, and financial loss. The paper focuses on safeguarding the system against the FDI (false data injection) attack on power system state measurement and estimation, which hinders the operator's insight of the system's actual operational status and prevents the operator from taking necessary countermeasures. Since FDI compromises the integrity of system state information, a Markov decision process (MDP) framework is proposed to model the strategy of FDI attack on state estimation in the power system, in order to assess the vulnerability of the power system to cyber-attack. Furthermore, a reinforcement learning (RL) paradigm is exploited to identify the optimal attack strategy, and the operator will solve the problem from the point of view of the intruder. The proposed framework is validated through numerical experiments conducted to investigate the optimal attack strategy with various test case scenarios.