Parameter-Free False Data Injection Attack Against AC State Estimation: A Canonical Polyadic Decomposition Based Approach

Abstract

With the evolving trend of modern power systems towards cyber-physical system (CPS), it is paramount to understand and investigate emerging threats, such as false data injection attacks (FDIAs). FDIA is capable of manipulating measurement data, thereby posing a serious risk to power systems. This paper proposes a new FDIA method against AC state estimation without the requirement on system parameters information. At first, the nonlinear AC state estimation model is formulated into a tensor form, where measuring variables are modelled as multiple tensor products between state variables and a third-order tensor characterising system information. Building upon this tensor-shaped modelling, measurement data is gathered into a diagonal tensor, following which tensor canonical polyadic (CP) decomposition is employed to factorize these data. The resultant lateral column space obtained by CP decomposition enables the stealth of the proposed FDIA method. In contrast to existing parameter-free FDIA methods in the literature, the proposed method makes no simplification for nonlinear AC model. Hence it is accurately consistent to the realistic power grid, and easier to bypass the bad data detection (BDD) of the target power grid. The proposed method is adaptive to the scenario that only data of partial sensors are available. Extensive simulation cases using synthetic data in numerous testing systems and comparisons with other parameter-free methods demonstrate the effectiveness and advantages of the proposed approach.