Data-Based Q-Learning for Replay Attack Detection in Cyber-Physical Systems

This article investigates an innovative data-driven approach to detect replay attacks in cyber-physical systems (CPSs). The core innovation lies in the pioneering application of a Q-learning algorithm for real-time estimation of the unknown system output values, which is integrated with measurement encoding techniques to transform replay attacks into additive disturbances in residuals for detection. This proposal effectively surpasses the limitation of traditional detection methods' reliance on system models while enhancing the sensitivity and accuracy of attack detection. Secondly, a replay attack detector based on residual analysis is constructed, and the detectability of replay attacks is theoretically proven. Finally, the effectiveness and superiority of the proposed scheme are verified through simulation and comparative experiments.