Nonlinear Granger causality graph method for data-driven target attack in power cyber-physical systems

Abstract

Owing to the deep integration of the information and communication technologies, power cyber-physical systems (CPSs) have become smart but are vulnerable to cyber attacks. To correctly assess the vulnerability of power CPSs and further study feasible countermeasures, we verify that a data-driven target attack on a nonlinear Granger causality graph (NGCG) can be constructed successfully, even if adversaries cannot acquire the configuration information of the systems. A NGCG is a unified framework for the processing and analysis of nonlinear measurement data or datasets and can be used to evaluate the significance of power nodes or lines. In addition, an algorithm including data-driven parameter estimation, noise removal and data reconstruction based on symplectic geometry is introduced to make the NGCG a parameter-free and noise-tolerant method. In particular, three new indexes on the weight analysis of the NGCG are defined to quantitatively evaluate the significance of power nodes or lines. Finally, several case studies of a nonlinear simulation model and power systems in detail verify the effectiveness and superiority of the proposed data-driven target attack. The results show the proposed target attack can select the key attack targets more accurately and lead to physical system collapse with the least number of attack steps.