Online Generative Adversary Network Based Measurement Recovery in False Data Injection Attacks: A Cyber-Physical Approach

Abstract

State estimation plays a critical role in maintaining operational stability of a power system, which is however vulnerable to attacks. False data injection (FDI) attacks can manipulate the state estimation results through tampering the measurement data. In this paper, a cyber-physical model is proposed to defend against FDI attacks. It judiciously integrates a physical model which captures ideal measurements, with a generative adversarial network (GAN) based data model which captures the deviations from ideal measurements. To improve computation efficiency of GAN, a new smooth training technique is developed, and an online adaptive window idea is explored to maintain the state estimation integrity in real time. The simulation results on IEEE 30-bus system and IEEE 118-bus system demonstrate that our defense technique can accurately recover the state estimation data manipulated by FDI attacks. The resulting recovered measurements are sufficiently close to the true measurements, with the error lower than $1.5e^{-5}$ and $2e^{-2}$ p.u. in terms of voltage amplitude and phase angle, respectively.