A Dversarial Attack for Deep Learning Model in Power System Transient Stability Analysis

The continuous application of advanced artificial intelligence technology takes a significant part in the management and control of power system. However, the machine learning model has the problem of adversarial examples, which may cause large deviation results that threaten the security of power system operation decision-making. Therefore, taking power system transient stability assessment (TSA) scenario as an example, a method of adversarial attack for deep learning model is presented in this paper. First, the method of constructing power system TSA model based on convolutional neural network (CNN) is introduced. Then, utilizing Carlini & Wagner (C&W) adversarial attack algorithm to generate adversarial examples and attack the power system TSA model. Finally, using IEEE 39 node system to verify the effectiveness of adversarial attack method for CNN-based power system TSA model.