Momentary ride comfort evaluation of high-speed trains based on feature selection and gated recurrent unit network

Abstract

With the rapid development of high-speed railways, momentary ride comfort related to temporary excitation is becoming an important concern of the corresponding technology development. This paper proposes a method for momentary ride comfort evaluation, where subjective and objective evaluations are related with each other by means of feature selection and gated recurrent unit (GRU) network. In terms of the concept of maximum information coefficient and the idea of max-relevance and min-redundancy, a feature selection algorithm, MIC-mRMR, is implemented for multi-channel selection. A data augmentation based on down-sampling and channel replacement is designed to enlarge the dataset size, and the short-time Fourier transform is adopted to effectively extract the time-frequency characteristics of the objective measurement signal. A GRU model is then constructed for subjective evaluation according to measured vibration data. Objective experiments and subjective evaluation were conducted on a specially designed comprehensive inspection train, where the proposed methods are applied and validated. It is thus demonstrated that the proposed methods are effective in evaluating the momentary ride comfort of high-speed trains.