Graph Autoencoder-Based Power Attacks Detection for Resilient Electrified Transportation Systems

The interdependence of power and electrified transportation systems introduces new challenges to the reliability and resilience of charging infrastructure. With the increasing prevalence of electric vehicles (EVs), power system attacks that can lower customers charging satisfaction rates are on the rise. The existing false data injection attacks (FDIAs) detection strategies are not suitable for protecting the power-dependent transportation infrastructure since (a) these detectors are primarily optimized for power grids alone, and (b) they overlook the impact of attacks on the quality-of-service of EVs and charging stations (CSs). In response to these challenges, this paper aims to develop an FDIA detection strategy that takes advantage of the data correlations between power and transportation systems, ultimately enhancing the charging satisfaction rate. To achieve this goal, we propose a graph autoencoder-based FDIA detection scheme capable of extracting spatio-temporal features from both power and transportation data. The input features of power systems are active and reactive power while those for transportation systems are the hourly traffic volume in CSs. The proposed model undergoes comprehensive training and testing on various types of FDIAs, showcasing improved generalization abilities. Simulations are conducted on the 2,000-bus power grid of the state of Texas, featuring 360 active CSs. Our investigations reveal an average detection rate of 98.3%, representing a substantial improvement of 15-25% compared to state-of-the-art detectors. This underscores the effectiveness of our proposed approach in addressing the unique challenges posed by power-dependent electrified transportation systems.