Graph Neural Network-Based Approach for Detecting False Data Injection Attacks on Voltage Stability

Abstract

The integration of information and communication technologies into modern power systems has contributed to enhanced efficiency, controllability, and voltage regulation. Concurrently, these technologies expose power systems to cyberattacks, which could lead to voltage instability and significant damage. Traditional false data injection attacks (FDIAs) detectors are inadequate in addressing cyberattacks on voltage regulation since a) they overlook such attacks within power grids and b) primarily rely on static thresholds and simple anomaly detection techniques, which cannot capture the complex interplay between voltage stability, cyberattacks, and defensive actions. To address the aforementioned challenges, this paper develops an FDIA detection approach that considers data falsification attacks on voltage regulation and enhances the voltage stability index. A graph autoencoder-based detector that is able to identify cyberattacks targeting voltage regulation is proposed. A bi-level optimization approach is put forward to concurrently optimize the objectives of both attackers and defenders in the context of voltage regulation. The proposed detector underwent rigorous training and testing across different kinds of attacks, demonstrating enhanced generalization performance in all situations. Simulations were performed on the Iberian power system topology, featuring 486 buses. The proposed model achieves 98.11% average detection rate, which represents a significant enhancement of 10-25% compared to the cutting-edge detectors. This provides strong evidence for the effectiveness of proposed strategy in tackling cyberattacks on voltage regulation.