Remaining Useful Life Prediction of High-speed Railway Traction System Based on Performance Degradation

To address the problems of complex structures, complicated performance degradation parameters, and missing failure threshold, we propose a remaining useful life (RUL) prediction method based on performance degradation. First, we extract various statistical features of degraded parameters, and perform feature selection using monotonic, related, and redundancy indexes to reduce the interference of redundant and irrelevant features. We propose a health-based strategy that assesses the system failure condition by assessing the health of the system to predict RUL without a failure threshold. Then, we use the selected features to train long- and short-term memory networks for degraded trajectory prediction. We conduct a case study using a hardware-in-the-loop simulation platform for the traction system of Chinese railway high-speed trains to predict the RUL of the DC-link circuit with capacitance degradation. Experimental results show the validity and superiority of the proposed method.