A CNN Based Transfer Learning Method for High Impedance Fault Detection

Abstract

High impedance fault (HIF) detection is a serious challenge of distribution networks. Data-based HIF detection methods have attracted the attention of researchers, but is hard to be applied into practical situation caused by unavailable data. In this paper, a Convolutional Neural Network (CNN) based transfer learning method for HIF detection is proposed by using distribution-level phasor measurement units (D-PMUs) data. The synchronous transient HIF characteristics are extracted of the zero sequence current by wavelet transform. To uniform the size of data, principal component analysis (PCA) is adopted to form the input feature. Then, a 3-layer CNN model was pre-trained by a typical 20-node distribution network. By fine-tuning with data augmentation, the pre-trained CNN model can be transferred to the target distribution network by just a small amount of data. The performance of proposed method was verified by 2 different distribution networks in PSCAD/EMTDC.