Because transformer plays an important role in power system, it rarely runs in fault or abnormal state. Therefore, it is difficult to obtain sufficient transformer fault samples. In this paper, aiming at the problem of less fault sample data and data imbalance, a novel data augmentation method based on kernel principal component analysis is proposed to non-linearly map the original data to a high-dimensional feature space. In this way, the new sample data retaining the feature information of the original data can be obtained. Second, the deep residual network is introduced with the identity path to construct the fault diagnosis model, which enables the weight parameters to be effectively transferred and updated. The simulation results show that the proposed method can effectively expand the data samples with high similarity with the original data, and the residual network model has strong feature extraction ability, which improves the accuracy of fault diagnosis. (c) 2023 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
