网络攻击下多智能体系统攻击检测设计与分布式弹性控制

This paper proposes an integrated strategy that combines attack detection and defensive control to investigate the resilient tracking control problem of multi-agent systems under actuator attacks. In terms of attack detection, a dynamic feature function modeling detection scheme based on state machine is designed. This scheme suggests employing a linear function observer to estimate the actuator signals and constructing an attack detection criterion based on the error characteristics between the actual signals and the estimated values, thereby enabling effective detection of actuator attacks. For the design of defensive strategies, to mitigate the impact of actuator attacks on the system＇s tracking consensus performance, this paper constructs a collaborative control system enhanced by a virtual network. By establishing connections with the leader and each follower, this system ensures resilient tracking control even when the information about actuator attacks is unknown. Specifically, when an attack is detected, the defensive control strategy of each follower switches to the control signal provided by the virtual layer; If no attack is detected, the original actual control signals of the followers are maintained. Compared with existing research findings, the controller designed in this paper does not require prior information such as the number of malicious nodes and the positions of attackers in advance, thus exhibiting stronger practicality and adaptability. Finally, a numerical example is presented to validate the proposed theoretical algorithm. The results demonstrate that the algorithm can effectively cope with actuator attacks and achieve resilient tracking control in multi-agent systems.