基于机器学习的信息物理系统安全控制

This paper investigates the security control problem of cyber-physical systems whose control signals are maliciously tampered. Firstly, a kernel extreme learning machine with improved fruit fly optimization (IFOA-KELM) algorithm is proposed to reconstruct the attack signal. Secondly, with the reconstructed signal treated as disturbance, a model predictive control strategy is designed to secure the system, and a condition that guarantees the input-to-state stability of the attacked system is given. In addition, to train the proposed algorithm, enough data of the system attacked with an optimal strategy is generated. This strategy is obtained by solving an optimization problem from the attacker's perspective. Finally, a numerical example of the spring-mass-damping system is illustrated to verify the effectiveness of the IFOA-KELM algorithm and the proposed control strategy.