Optimal Node Attack on Causality Analysis in Cyber-Physical Systems: A Data-Driven Approach

Abstract

This paper focuses on the data-driven optimal attack strategy against state estimation in cyber-physical systems (CPSs). Different from the research on attack strategies of specific attack types, the proposed attack strategy addresses the optimal selection of attacked targets, which can combine with different attack types and produce greater threats to CPS. In particular, a causality analysis (CA) on the measurement data is first proposed to evaluate the significance of nodes (sensor groups) and help the implementation of the optimal node attack, since the system topology and parameters are not available to adversaries. On the one hand, a multivariate transfer entropy and several data preprocessing methods are employed to complete the CA between sensor groups qualitatively. On the other hand, three new indexes, e.g., driver degree, are defined to complete the CA quantitatively. Moreover, the theoretical basis for the proposed node attack is provided, in which the superiority of the node attack is proven from the view of observability. Finally, the case studies on the smart grid are illustrated to verify the superiority of the proposed attack strategy.