A Power System Data-Driven State Estimation Adversarial Attack Method Based on Conditional Generative Adversarial Network

EasyChair Preprint 15370

Abstract

Data-driven methods based on artificial intelligence technology exhibit the ability to process data quickly and accurately, thus currently being widely applied in the field of power system state estimation (SE). However, recent studies have found that processing data during the training phase of data-driven algorithms can mislead the operational results, demonstrating security risks in data-driven methods. In light of this, to reveal potential security issues in data-driven algorithms, this paper proposes an adversarial attack method based on conditional generative adversarial network (CGAN). Experimental results indicate that adding minute disturbances generated by CGAN to target samples can misleads the predictions of the power system SE model. This method enables directing the attack according to preset label conditions during the model application phase.

Keyphrases: Adversarial Attack, conditional generative adversarial network, data-driven algorithms, state estimation

@booklet{EasyChair:15370,
  author    = {Shiyi Hou and Jing Zhang and Lei Zhu and Qi Wang},
  title     = {A Power System Data-Driven State Estimation Adversarial Attack Method Based on Conditional Generative Adversarial Network},
  howpublished = {EasyChair Preprint 15370},
  year      = {EasyChair, 2024}}