Data-driven RUL Prediction of High-speed Railway Traction System Based on Similarity of Degradation Feature

Abstract

The remaining useful life (RUL) prediction of high-speed railway traction system is of great significance for ensuring the safe and efficient driving of high-speed railway trains. Due to the complex structure of high-speed railway traction system, it is difficult to reveal system-level degradation mechanism; thus, a data-driven RUL prediction method based on similarity of degradation features is proposed in this paper. The seq2seq structure of the Long Short Term Memory (LSTM) is adopted to extract the multivariate features of the degradation trajectory. Based on these features, a similarity-based RUL prediction method is utilized to compute the RUL of the system. Experiments are conducted on the semi-physical platform of the CRH2 traction system. Results can show that the proposed method can extract reasonable degradation features; and the prediction accuracy is greatly improved compared with several existing methods.