Resilient Cooperative Control Of Heterogeneous Multi-agent Systems Against Cyber Attacks

In the current era of emerging technologies,distributed multi-agent systems have developed into a highly anticipated research field due to their wide range of potential applications in engineering and defense.In the complicated real-world environments,such as coordination of air,ground and sea,power system stability control,and multi-satellite attitude alignment and so on,multi agents exhibit heterogeneous characteristics,and they communicate with each other by open networks to achieve collaborative control.However,the convenient means of communication have rendered the system vulnerable to cyber attacks.Thus,to prevent malicious attack signals from spreading through the network,it is crucial to construct resilient distributed cooperative control strategies for the security improvement of heterogeneous multi-agent systems to improve.This paper focuses on the concerns of the heterogeneous multi-agents systems subject to typical cyber attacks.By compensation,isolation and filtering methods,the adverse effects of attack signals on the system can be alleviated or eliminated.In this regard,multiple resilient cooperative control strategies are provided to enhanced the security and reliability of the system based on the graph theory,L₂ gain theorem,and output regulation theory.The contributions of this paper include the following three folds:1 This paper proposes a compensation-based resilient control strategy for heterogeneous multi-agent systems that state information suffer from unbounded sensor attacks.Firstly,for the heterogeneous multi-agent system,a secure distributed state observer that independent of network physical layer is constructed.To estimate the unmeasurable actual state of the system and the unknown sensor attack signal,respectively,a resilient observer and a sensor attack signal estimator are designed together to establishL₂ gain design.Then,consider the control signals under bounded actuators attacks,an adaptive actuator attack compensator signal is designed using the state estimation.Finally,from Lyapunov function,the closed-loop system is proven to be ultimately uniformly bounded,that is to say that the proposed control strategy drives the followers into a small neighborhood of the convex hull spanned by the outputs of the leaders.2 This paper proposes an isolation-based resilient control strategy for heterogeneous multi-agent systems that output information suffer from sensor attacks.Firstly,a virtual layer with the same output matrix is designed for each follower combined with the output regulation equations,and an attack monitor based on relative output information from virtual layer is designed to detect misbehavior of neighboring agents.Then,a multi-round attack detection and isolation algorithm is proposed to protect normal followers from being misdiagnosed and remove the compromised followers from the communication topology.Finally,design a control protocol that only uses relative output information for normal followers,and the closed-loop system is proven to be stable byL₂ gain theory.3 For linear high-order heterogeneous multi-agent systems subject to f-local/total Byzantine attacks,a resilient control strategy based on censoring mechanism is proposed.Firstly,two structurally balanced bipartite network topologies with strongly robustness are designed to provide graph-theoretic analysis for two kinds of attack models.Secondly,a weighted bipartite mean-subsequence-reduced（WB-MSR）algorithm is proposed for the bipartite network topologies with competitive/cooperative relationships among followers.The true state values are multiplied by the values that the weight coefficients among followers are processed by signum functions,and then the algorithm has to sort and censor the product result.Then,a resilient adaptive observer is designed without any global topological information,whose input parameters are the filtered values obtained by separately executing the WB-MSR algorithm for each state component.Meanwhile,the observer is able to provide secure estimation of the leader’s state.Finally,the exponentially convergent bipartite output consensus is achieved by utilizing a distributed feedforward control method.