Estimation of Vertical Track Irregularity Based on NARX Neural Network

Abstract

A method based on NARX neural network was proposed to estimate the vertical track irregularities. First, the basic theory of track vertical irregularity and inertial methods used to monitor track irregularity were introduced respectively. Second, NARX neural network structure and Bayesian regularization network training algorithm were described in detail. Third, a vehicle-track dynamic model was developed. Then, the actual track irregularity data from high-speed line was used to obtain the simulation response data. Finally, the normalized data was used as input of the NARX model and the data of track irregularity was used as output. The root-mean-square error (RMSE) and correlation coefficients were applied to evaluate the network performance. The experimental results show the efficiency and accuracy of the presented method to assess the vertical track irregularities.