Data-Driven Stealthy Attacks Detection in Cyber-Physical Systems Based on Complex Dynamical Networks Encoding Strategy

This article develops a novel data-driven stealthy attack detection strategy for a class of cyber-physical systems (CPSs) with external disturbances. First, it is proved that there exists a class of data-driven stealthy attacks designed by the subspace identification method, which cannot be detected by the existing detection protocols including the constant matrix-based encoding strategy. Then, in order to overcome this difficulty, the complex dynamical networks (CDNs) and the encoding/decoding technique are introduced to detect the data-driven stealthy attacks. In particular, the synchronization technique is adopted to ensure the consistency of the key sequences in encoding/decoding process, so that the encoded information on the decoder can be correctly recovered without attacks. In addition, the case of information leakage is analyzed, and it is demonstrated that the existing encoding detection strategy based on single node chaotic systems is ineffective, while the proposed one enhances the complexity of the encoding link and can still distinguish the stealthy attacks. In the end, simulations for the model of a DC motor system are performed to verify the effectiveness of the presented CDNs-based encoding detection scheme.