Stealth FDIA Localization in Power Systems Using Spatio-Temporal Graph Neural Networks

Stealth false data injection attacks (FDIA) pose enormous threats to cyber-physical power systems. These attacks exhibit spatial stealthiness, making it challenging for existing bad data detection to accurately locate them. In this paper, a novel Spatio-Temporal Graph Neural Networks (STGNN) model is proposed to detect and locate FDIA accurately. First, FDIA is proposed to be generated by regional state estimation where only measurements and topology parameters specific to the attacked area are required. Then, state estimation is performed to convert line flow signals into node state signals to reduce the impact of noise with redundant measurement information. Finally, STGNN is proposed to extract the temporal and spatial features. The proposed model is evaluated on an IEEE 14-bus system. The simulation results demonstrate that the proposed method achieves high-resolution localization of FDIA leveraging the STGNN model.