Geometry-Based Data-Driven Complete Stealthy Attacks Against Cyber-Physical Systems

Abstract

This paper proposes a data-driven complete stealthy attack strategy against cyber-physical systems (CPSs) based on the geometric approach. The attacker aims to degrade estimation performance and maintain stealthiness by compromising partial communication links of the actuator and sensor. Different from the classic analysis methods that require accurate model parameters, we focus on how to establish the connection between geometry and data-driven approaches to represent the malicious behavior of attacks on state estimation. First of all, the existence of complete stealthy attacks is analyzed. Then, the maximal attached stealthy subspace and the set of estimation errors under complete stealthy attacks are analyzed intuitively from the geometric point of view. On this basis, the complete stealthy subspace is constructed with the subspace identification method, which is applied to generate the corresponding stealthy attack sequence through the collected system input-output data. Finally, simulation results are provided to illustrate the effectiveness of the proposed strategies.