Deep Learning -Based Remaining Life Prediction for DC-Link Capacitor in High Speed Train

The accurate prediction of the remaining useful life of DC-link capacitors is crucial in high-speed railway traction drive systems. In this paper, we propose a method for predicting residual life using deep learning by particle swarm optimization. Firstly, the upper and lower voltage values of the capacitor are selected as the characteristic values. By applying wavelet transform to denoise the voltage, we can calculate failure thresholds. Then, we amplify the input weights in real-time using the long short-term memory neural network with a macro-micro attention mechanism. Finally, we utilize the particle swarm optimization algorithm to optimize the number of input units and the learning rate of the neural network for the purpose of predicting lifetime. The effectiveness of this method is verified through model evaluation indices in a case study of high-speed train traction system.