A Deep Reinforcement Learning Framework for Attack-Fault Tree Analysis and Security Assessment in Power Systems

Cyber-physical systems (CPS), such as power plants and critical infrastructure, face growing safety and security risks due to increased interconnectivity and automation. To address these challenges, we propose a framework that combines deep reinforcement learning (DRL) and Bayesian probability to model and assess vulnerabilities and attack paths. Leveraging Deep Q-Learning Networks (DQN) and cumulative probability, our method improves the identification and prioritization of effective attack paths. We also introduce the attack-fault tree (AFT) model to evaluate interactions between safety and security events. Using Uppaal SMC and statistical timed automata, we simulate dynamic scenarios and generate probabilistic estimates of metrics such as attack cost, duration, and impact. By analyzing both current and hypothetical scenarios, our approach supports informed, adaptive defense strategies for resilient CPS.