Detection of Cyber Attacks on Leader-Following Multi-Agent Systems

This paper studies an attack detection problem for a networked leader-following multi-agent system subject to unknown-but-bounded system noises and quantization effects, where an adversary launches malicious cyber attacks on agents' measurement outputs aiming to distrust the leader-following consensus. An effective distributed attack detection algorithm is firstly developed for each follower such that the attack can be identified at the time of its occurrence. The core of the algorithm lies in a set-membership filtering approach from which each designed filter can provide an ellipsoidal state prediction set and an ellipsoidal state estimation set. Whether a filter can detect the occurrence of such an attack is then determined by the existence of intersection between these two sets. Furthermore, a convex optimization algorithm is established to solve out anticipated consensus protocol and two-step set-membership filter by resorting to some recursive linear matrix inequalities. Finally, an illustrative example is given to show the effectiveness of the proposed main results.