Generalized Graph-Dependent Isolation of Collusive Attacks for Interconnected Systems

• 成果类型: Article
• 署名作者: Zhao, Dan; Ho, Daniel W. C.; Wen, Guanghui
• 署名单位: Southeast University - China; City University of Hong Kong
• 刊物名称: IEEE TRANSACTIONS ON AUTOMATIC CONTROL
• ISSN/ISSBN: 0018-9286
• DOI: 10.1109/TAC.2024.3474055
• 发表日期: 2025
• 关键词: DISTRIBUTED FAULT-DETECTION; large-scale systems; resilient control; TOLERANT CONTROL; consensus

摘要

This article addresses the isolation problem of collusive false data injection attacks and covert attacks for interconnected systems, particularly in establishing generalized graph-dependent conditions responsible for attack isolation. Two algorithms involving inclusive neighbor information and neighbor information are introduced to accomplish attack isolation. By exploring the mechanism of attack collusion, it is found that both the number of subsystems' one-hop neighbors and the size of cycles in the graph play significant roles in attack isolation. Besides, it is interesting to note that their impacts on attack isolation can be mutually compensated. Then, the worst attack collusion cases for the proposed algorithms under different cycle sizes are developed, based on which, several equivalent sufficient yet sharp graph conditions for the isolation of false data injection attacks and covert attacks are, respectively, provided with the notion of $(r,s)$-isolability. It is revealed that the inclusive neighbor information-based algorithm is more effective for isolating false data injection attacks. In contrast, the neighbor information-based algorithm is more suitable for isolating covert attacks. Finally, simulations demonstrate the effectiveness of the theoretical results.