Time-frequency Fusion Method via Convolutional Neural Network for Partial Discharge Classification

Abstract To improve the accuracy of partial discharge (PD) pattern recognition by jointing time-domain (TD) and frequency-domain (FD) information, a time-frequency (TF) fusion method via convolution neural network (CNN) is proposed in this paper. Firstly, PD signals are represented waveform images transformed into envelope variational mode decomposition-based Hilbert marginal spectrum (VHMS). Secondly, network, FuNet involving 2-dimensional CNN (2D-CNN), 1D-CNN, multilayer perceptron (MLP), established to join TF information. In FuNet, 2D-CNN inputted 1D-CNN adopted VHMS signal as its input all improved drawing on complementary strengths different layers’ features. Then MLP will fuse extracted TD FD features classify defects.