Transient Fault Signal Identification of AT Traction Network Based on Improved HHT and LSTM Neural Network Algorithm

This paper aims to address the difficult pinpoint fault cause of full parallel AT traction power supply system with special structure. The characteristics are easily covered up, and high transition impedance only affects singularity wavehead, making traveling waves hard identify. Moreover, classification accuracy traditional time-frequency analysis method is not sufficiently distinguish precisely. In this paper, a network based on single-channel improved Hilbert–Huang transform deep learning proposed. extracts effective features directly from original signals classifies types at same time. data categorization increased by applying extract transient produce one-dimensional feature data, which analyzed energy spectrum. Using similarity recognition long-short-term memory neural network, extracted high-frequency trained tested classify more accurately. order verify effectiveness method, several kinds short-circuit lightning strike faults continuously simulated verified in paper. Considering various conditions factors, proposed HHT+LSTM compared LSTM for direct processing signals. HHT + algorithm achieves an 99.99%.