Low-Rank Undetectable Attacks Against Multiagent Systems: A Data-Driven Approach

Abstract

This article is concerned with the problem of potential security threats in multiagent systems (MASs) with consensus control protocols. Most existing security strategies focus on resilient control schemes, ignoring potentially more stealthy malicious attacks. To explore system vulnerabilities, a class of data-driven low-rank undetectable attack strategies is investigated against communication channels in MASs. The objective of attacks is to corrupt minimal sensors to degrade the state estimate performance while avoiding being detected. First, the conditions for the existence of low-rank undetectable attacks are determined by analyzing the kernel space of the low-rank subspace of the system expansion matrix. Utilizing the accessible measurement data, the attack matrix under undetectable conditions is constructed using the subspace identification method. To avoid proposing attack vectors on the full column space of the attack matrix, a bilateral random projection algorithm is designed to derive a low-rank approximation of the attack matrix. By analyzing the kernel space of low-rank undetectable attacks, an undetectable attack sequence is generated. Simulation results validate the effectiveness of the proposed low-rank undetectable attack algorithm.