A Fault Pattern and Convolutional Neural Network Based Single-phase Earth Fault Identification Method for Distribution Network

Abstract

The fault current is small when a single-phase earth fault occurs in distribution network, so it is still difficult to diagnose the faults accurately. To improve the identification accuracy of single-phase earth faults in distribution network, a fault pattern and convolutional neural network (CNN) based identification method is proposed in this paper. Firstly, this paper extracts the features of different fault types, and a classifier is constructed to identify the single-phase earth fault with low resistance fault and other faults. Then, the Hilbert-Huang transform (HHT) is used to transform the features of special faults from the time-domain to the frequency domain information. Finally, a CNN based classifier is proposed to distinguish the single-phase earth faults with high fault resistance and intermittent characteristics. To investigate the performance of the proposed identification method, simulations of a distribution grid case are carried out in MATLAB. Compared with artificial neural network (ANN) and k-nearest neighbor (KNN), the simulation results demonstrate that the proposed method can achieve higher accuracy on single-phase earth faults.