Analysis and Research of Neural Network Adversarial Samples for Power Grid Security
Bokai Li, Shenglong Wu, Yu-Ke Yang, Geyao Zhang · 2023
In the electric power information system security attack and defense in the industry, the use of big data adversarial samples to attack the neural network model can enhance the authenticity of the adversarial samples. This paper strengthens the research on the authenticity of the samples by analyzing and studying the generation of neural network adversarial samples and simulated at-tack tests. Through a variety of experimental methods, for the improvement of complex neural network adversarial samples, the generation method of semantic adversarial samples is improved and perfected, and adversarial disturbances are added. The experimental results show that after ComDefend and feature compression defense processing, the authenticity is improved by at least 30% compared with the traditional method.