On Attack Detection With Optimal Policy Based On The Theory Of Discrete Event Systems

As a new concept to integrate sensing,computation,control and networking in-to physical objects and infrastructure,cyber-physical system(CPS)has been widely used in daily life and industrial production.However,its computing characteristics and openness make it vulnerable to external attacks.Especially in the critical infras-tructure,the security of its control system is particularly important.Based on such a requirement,this paper studies the distributed multi-attack detection problem of partially-observed discrete event system,where the system is monitored by a group of agents,which can control the on-off states of sensors to cooperatively detect a set of attacks within finite steps.The objective of this paper is to find a distributed sensor activation policy for agents to detect all attacks,and then propose a method to optimize the policy.For detection problems,we first discuss the attack model,establish the structure of observer and introduce the concept of detectability.Then the state classified function is established based on the finite counting automata,and together with the state-partition automaton introduced,the detection function can be constructed to complete the detection work.Next,to solve the optimization prob-lem,the initial problem is preliminarily represented as a distributed minimization problem.By extending the person-to-person approach and applying the detection function again,the distributed minimization problem can be decomposed into several centralized minimization problems.Finally,a novel synthesis algorithm is proposed to solve these centralized minimization problems in turn by following the structure of the most permissive observer and adopting a greedy policy,and the optimal policy can be obtained by executing the operation of pruning.Theoretical proof shows that the proposed algorithm can generate language-based optimal observer to detect all attacks.