False Data Injection Attacks Detection in Wide Area Damping Control System Using Semi-Supervised Generative Adversarial Network Model

The drive towards digital automation of the power grid has resulted in the extensive deployment of wide-area measurement system (WAMS) technology. This WAMS technology is communication network-based and responsible for transmitting measurements and control signals to enable the operation of wide-area damping controllers (WADC). However, the inherent vulnerability of communication networks exposes the power system to various cyber threats, such as false data injection (FDI) attacks, delay attacks, and denial of service (DOS) attacks. Although researchers have developed numerous supervised machine learning and deep learning solutions, but the unavailability of labeled large attack data sets in the real world presents a significant challenge for detecting cyber attacks. To address this issue, a novel approach is proposed that utilizes semi-supervised generative adversarial networks (SSGAN) in conjunction with physics-aware features for FDI attack detection. The proposed framework uses the damping torque coefficient (DTC) and mode shape (MS) features for training, which accurately captures the system dynamics during event and attack scenarios. Various FDI attacks and power system events are considered in this study. To validate the proposed method, tests are conducted using the 4-machine Kundur’s system. The results demonstrate that the proposed framework significantly enhances the cybersecurity of the power system’s cyber layer.