Anomaly data enhancement and detection for charging piles considering generative adversarial network
Q. Feng, Yaoyu Zhou, Haizhou Li, Dan Feng · IET conference proceedings. · 2023
The accuracy of anomaly detection results is affected by factors such as low proportion of abnormal data samples and imbalanced distribution. To address the issue that the current abnormal data detection model for charging piles depends on the quality of abnormal data samples in the training set, this paper proposes a charging pile abnormal data augmentation method based on Generative Adversarial Network (GAN) and the corresponding abnormal data detection model. The proposed method first improves the traditional GAN by introducing the unsaturated activation function Leaky ReLU (LR) and proposes charging pile abnormal data augmentation method based on LR-GAN to enhance the abnormal data in the training set. Then, the LR-GAN augmented training set is used as input for the Random Forests (RF) model to establish a charging pile abnormal data detection model based on LR-GAN-RF to detect abnormal data in the charging pile charging dataset. Finally, different charging pile anomaly data sets are used to validate the feasibility and effectiveness of this method, which is verified to be highly feasible.