A Transformer Approach to Identifying False Data Injection Attacks
Hu Li, Huan Pan, Jiayi Jin, Mengna Sun · 2024
The smart grid is vulnerable to multiple cyberattacks, with false data injection attacks (FDIA) being particularly concerning, as they can evade conventional bad data detection (BDD) systems and disrupt power grid operations. Given the nonlinear nature of state estimation and grid data, along with the temporal continuity of state data, this study introduces a Transformer-based model for detecting FDIA. It utilizes the Hilbert-Huang Transform (HHT) to capture precise time-frequency characteristics from power system time-series data. These features are input into the Transformer, where multi-head attention captures long-range dependencies and improves recognition of complex grid patterns. The application of this technique on the IEEE 14-bus system demonstrates its capability to effectively recognize abnormal patterns with high accuracy. In addition, a basic sensitivity analysis of the control parameters was undertaken.