A GAN-based Adversarial Attack Method for Data-driven State Estimation

With the development of data science and technology, data-driven methods can provide support to real-time and accurate state estimation (SE) of power systems. However, the non-interpretability and poor robustness of data-driven algorithms also introduced new security risks to the power system. The risk transmitted to the related operations of state estimation can cause errors in system state perception, or even lead to more serious consequences. In this paper, an adversarial attack method for data-driven state estimation based on generative adversarial network (GAN) is proposed, which aims at revealing and uncovering the security risks of data-driven SE algorithms. Simulation results show that the GAN-based adversarial attack can cause significant deviations of power system state estimation, which can lead to subsequent security problems.