Attacks Allocation And Defense Design For Multi-agent Systems

The multi-agent systems have a variety of potential applications in different fields such as intelligent transportation systems,surveillance and reconnaissance systems,formation flying of unmanned aerial vehicles and so on.The communication topology plays an essential role in the multi-agent systems.Unfortunately,some malicious attacks can destroy the system by occupying the communication resources of the multi-agent systems(MASs)or compromising the communication data.The typical forms of attack include denial of service(Do S)attacks and false data injection(FDI)attacks.The former indirectly changes the communication topology of MASs,while the latter can compromise the communication data.When studying the defense strategy of MASs,it is necessary to analyze the influence of network attacks on MASs firstly.Hence,the problems involved in this dissertation are divided into two parts.Firstly,from the attacker’s point of view,policies of Do S attacks and completely stealthy FDI attacks are designed respectively.Secondly,from the point of view of defender,secure defense mechanisms are designed for sensor network and leader-following system to against stealthy FDI.The main contents of the dissertation are summarized as follows:1.For the leader-follower multi-agent systems,the scheduling mechanism of Do S attacks with minimum energy is designed to destroying the consensus of system.Firstly,the problem of destroying the global reachability of topology is transformed into the minimum cut of a flow network.Then an algorithm of generating the minimum cut of flow network is designed.Finally,the upper and lower bounds of attack probability are given based on the stability theory of stochastic systems.2.A general FDI attack model with energy restricted is designed for a sensor network system with remote estimator.Firstly,the probability distribution function is adopted to describe the stealthiness of attack which differs from existing works.Then the relationship between the model and the existing works is analyzed.Finally,the minimum number of compromised sensors is given based on the spanning space of related eigenvectors.3.The completely stealthy FDI attacks model obtained aforementioned is extended to MASs.The attacks policy is designed under the situation of energy limitation.Differ from the situation of central system,the topological condition of the complete stealthiness is transformed into the solution of Riccati equation by introducing an exogenous signals.Then,the problem of minimizing compromised communication edges is replaced by a minimum cut of related flow network.Finally,an algorithm is provided to generate a completely stealthy FDI attacks with minimum attack targets.4.To defence the stealthy FDI attacks aforementioned,the secure defence strategies are designed.Firstly,a stochastic data transmission mechanism is given for the sensor network system with remote estimator.By introducing the upper and lower bound of the estimation error covariance,the normal working conditions of remote estimator are given.Secondly,the defense strategy previous is extended to the leader-follower system.Meanwhile,the influence of defense strategy on the parameter of distributed controller is analyzed.