Attack Isolation And Resilient Control Of Multi-agent Systems

With the development of Internet of Things technology,the physical systems and communication networks have carried out with deep cooperation and organic integration,and the individuals with certain intelligence have evolved into all kinds of networked multi-agent systems(MASs)through communication cooperation and mutual coupling.However,the open structure of communication network and more intelligent hacker technology bring more abundant infiltration paths,stronger invisibility and greater destruction for attackers to invade the MASs,which not only brings great loss to social production and life,but also seriously threatens the security of country and people’s life and possessions.Therefore,the research on the secure control of MASs is both an important measure to speed up the construction of Internet Power,but also a significant way to implement the holistic approach to national security and realize the longterm stability of the country.Based on this,this paper studies the attack isolation and resilient cooperative control of MASs under collusive attacks,and explores the relationship between the communication topology and the attack-isolation-based resilient control mechanism,to build highly robust and invulnerable MASs.The main content includes the following four aspects.For the MASs subject to physical attacks,by using the distributed fixed-time observer,the resilient consensus of MASs is achieved based on attack isolation.First,the residual between the consensus error of each agent and its observation is derived,and the locations of the attacks are limited into the agent and its neighbors.Then,the distributed attack isolation algorithm is developed based on the non-pattern of the residuals of the agent and its neighbors.Besides,the concept of r-isolability is proposed to avoid mistaken isolation of normal agents,and thus,the proposed algorithm with r-isolability realize the attack isolation with zero missing alarm and zero false alarm.Finally,to avoid the influence of attacked neighbors on normal agents,the attack-isolation-based control algorithm is designed to achieve resilient consensus.For the MASs subject to collusive attacks,the residuals of certain agents are not able to detect whether there are attacked agents in the subnetwork consisting of itself and its neighbor,and the resilient consensus problem of MASs under collusive attacks is addressed.The stack vector consisting of the residuals of each agent and its neighbors is constructed,and the distributed attack isolation algorithm is developed based on the L0-norm of the stack vectors,where the isolated agents help to decouple the collusion among attacks and facilitate the isolation of other attacked agents.Furthermore,to ensure that all the attacked agents are isolated,the concept of(r,s)-isolability is introduced for the success of the attack isolation algorithm.Finally,the resilient control algorithm against collusive attacks is proposed to achieve the consensus of normal agents even if there are some normal agents being isolated mistakenly.For the MASs subject to collusive covert attacks,the covert attack on the agent cannot be detected by the agent itself,and the resilient stability problem of heterogeneous MASs is investigated.A two-stage fixed-time observer is built to derive the residual between the output of each agent and its observation,and the residual of each agent is used to detect whether there are covert attacks in its neighbors.Then,the distributed attack isolation algorithm against non-collusive covert attacks is derived.With the cooperation of the concept of r*-isolability,the attack isolation with zero missing alarm and zero false alarm is accomplished.Moreover,the stack vector consisting of the residuals of the neighbors of each agent is established,and the distributed attack isolation algorithm against collusive covert attacks is developed based on the L0-norm of the stack vectors.Finally,the resilient control algorithm based on attack isolation is derived to reach the stability of some normal agents.For the MASs subject to collusive attacks on communication links,the problem of resilient consensus based on attack isolation is studied.First,the distributed fixed-time observer is used to derive the residual between the output consensus error of each agent and its observation,and the locations of the link attacks are constrained on the agents involved with the links.Then,the distributed attack isolation algorithm against non-collusive link attacks is proposed based on the residuals of the agents on the links,and the topology condition is provided to ensure the correct isolation of attacked links and no mistaken isolation of normal links.Then,the distributed attack isolation algorithm against collusive link attacks is designed to decouple the collusion and isolate the link attacks.Finally,the distributed attack isolation algorithm and trusted links are cooperated to help achieve the leader-follower resilient consensus with relaxed topology conditions.