Attack Analysis for Discrete-time Distributed Multi-Agent Systems

Abstract

The recent growth of cyber-physical systems provides new opportunities to the attackers to undermine the system performance or even destabilize it. This work presents a system theoretic approach for the analysis of the adverse effects of attacks in discrete-time distributed multi-agent systems. Analyzing cyber-physical attacks from the attacker’s perspective improves the system awareness against attacks and reinforces the necessity of developing novel resilient control protocols. To this end, we first show that an attack on a compromised agent can adversely affect intact agents that are reachable from it. Then, we show that how an attack on a single root node can snowball into a network-wide attack and even destabilize the entire system. Finally, we show that the local neighborhood tracking error for the intact agents goes to zero, despite attack. This makes existing robust control approaches that aim at attenuation of disturbance or attack based on local neighborhood tracking ineffective, and demands designing novel resilient control approaches.