Physics-Aware Watermarking Embedded in Unknown Input Observers for False Data Injection Attack Detection in Cyber-Physical Microgrids

• Published in: IEEE transactions on information forensics and security 2024, Vol.19, p.7824-7840
• Main Authors: Liu, Mengxiang, Zhang, Xin, Zhu, Hengye, Zhang, Zhenyong, Deng, Ruilong
• Format: Article
• Language: English
• Subjects: cyber-physical co-simulation; Degradation; False data injection attack; microgrid; Microgrids; Monitoring; Perturbation methods; physics-aware watermarking; proactive detection; Real-time systems; Robustness; unknown input observer; Watermarking
• ISSN: 1556-6013; 1556-6021
• DOI: 10.1109/TIFS.2024.3447235

The physics-aware watermarking-based detection method has shown great potential in detecting stealthy False Data Injection Attacks (FDIAs) by adding appropriate watermarks to control commands or sensor measurements, especially in industrial control systems and grid-tied Distributed Energy Resources (DERs). However, existing watermarking-based detection methods have limitations in either handling the intricate physical couplings among DERs or characterising the fast changing power electronics dynamics, and thus cannot be directly applied to microgrids. Inspired by the methodology of Unknown Input Observer (UIO), which can be employed for the distributed anomaly monitoring in cyber-physical microgrids but would be easily bypassed once the adversary has the knowledge of certain electrical parameters, this paper makes the first attempt to investigate the physics-aware watermarking embedded in UIOs such that the stealthy FDIAs would be intentionally disrupted by the watermarking scheme. Based on the theoretical analysis of the detection enhancement and performance degradation under watermarking-enhanced UIOs, the watermark strengths, UIO parameters, and control gains are optimally co-designed to significantly enhance the detection effectiveness while not degrading the control performance. The robustness of the watermarking-enhanced UIO to Time Synchronisation Errors (TSEs) is improved by employing a sliding time window with appropriate length. The performance of the proposed method is validated through Matlab/Simulink studies and cyber-physical co-simulation experiments, and the sensitivities of the detection latency and TSE robustness to watermark strength and detection window's length are comprehensively studied.