Data augmentation for fault diagnosis of oil-immersed power transformer

Abstract

110 kV oil immersed transformer is a key part of the power transmission and transformation system, which determines the power quality and transmission efficiency. Its fault diagnosis can greatly reduce the maintenance cost and improve the economy. At present, the methods of transformer fault diagnosis have a strong dependence on the original data, and the size of the original data directly affects the effect of fault diagnosis. In order to change this situation and achieve higher accuracy of transformer fault diagnosis, this paper firstly uses the Conditional Variational Automatic Encoder (CVAE) composed of full connection layers to expand the original samples under each fault category. After data augmentation, the convolutional neural network (CNN) with strong feature extraction ability is selected as the classifier. Finally, the CVAE-CNN model is validated using public dataset and the result is compared to other machine learning algorithms.