Deep Domain Adaptation for Powe Transformer Fault Diagnosis Based on Transfer Convolutional Neural Network

Abstract

The power transformers are important devices in power systems. Some issues still exist and have been well addressed in the traditional methods. The traditional data-driven methods use the training data samples to train the samples and ignore the data distribution differences. This decreases the classification performance of the trained model on the testing data set. To address this problem, we proposed a transfer convolutional neural network (TCNN), which considers both of the classification loss on the domain data samples and the domain transfer loss. In this way, the proposed model has higher transferability and generalization ability on the testing samples, and thus the classification performance has been improved. tal results validate the effectiveness of the proposed method.