A Data-Based Moving Target Defense Framework for Switching Zero-Sum Games

In this paper, a data-based moving target defense framework for cyber-physical systems evolving with unknown and discrete-time dynamics is proposed. Specifically, we develop a proactive mechanism to increase the attacking surface through entropy-based unpredictability measures, and a reactive mechanism to detect and mitigate sensor/actuator attacks. In order to handle worst-case disturbances, we formulate our problem as a zero-sum game, where the minimizing player is the control input and the maximizing player is the disturbance input. We amalgamate a model-free and data-based approximate dynamic programming technique that learns the saddle-point strategies with a Bellman-based intrusion detection mechanism. Switching rules that asymptotically stabilize the switched system are derived. We validate the effectiveness of our proposed framework through simulation results.