Link false data injection attacks on multi-agent systems with attacked target nodes

This paper investigates the vulnerability of multi-agent systems under link false data injection attacks, where adversaries disrupt the leader-following consensus by tampering with the transmitted data between neighbor nodes. In contrast to existing network attack approaches that aim to destabilize at least one or all nodes, we consider a more general attack scenario where only the target nodes are compromised, while the remaining nodes remain healthy. Recognizing that attacks intend to maintain stealthiness as much as possible, an improved stealthiness constraint for attacks is introduced, which requires that all neighbor error increments induced by attacks converge to zero after a finite period. By analyzing the spectral radius of the system matrix and the reachability of its eigenvectors, we establish some necessary and sufficient conditions for the existence of attacks. To mitigate the attack impact, a channel protection strategy is proposed, which remains effective even if the target nodes of attackers are unidentified. Using the secure data from the protected channels, the leader-following consensus of multi-agent systems is successfully achieved. Simulations are provided to demonstrate the effectiveness of the presented theoretical results.