Zero-Sum Optimal Control for a Cyber–Physical System in Unreliable Communications via Secure Decentralized Policy Iteration

This paper investigates the robust consensus scheme for a specific type of leader-follower multi-agent systems (MASs) in the context of a human-cyber-physical interactive system. The system encounters frequent communication challenges characterized by packet loss and data leakage. To address these challenges, the paper proposes a secure decentralized scheme that leverages game theory to determine optimal policies in a zero-sum game. The scheme utilizes the adaptive dynamic programming (ADP) method, which involves a decentralized iterative process. The paper demonstrates that the generated sequence exhibits exponential convergence and provides the maximum iteration number based on the convergence error. To mitigate data leakage among players, the scheme incorporates a secure operation that involves encrypted data. This integration seamlessly combines data conversion and encryption-decryption into decentralized computation. Finally, the paper presents a simulation example to validate the efficacy of the consensus scheme.