Prediction of Ride Comfort Values Based on GA-BP Neural Network

Abstract

Various standards for evaluating ride comfort of railway vehicles have been developed and used in many countries. In order to construct the mapping relationships of several ride comfort standards, prediction of ride comfort values had been studied. Firstly, track irregularities were generated by track spectrum, and were imported to ADAMS/Rail to obtain vibration accelerations of carbody. Secondly, ride comfort values were calculated according to the algorithm of ride comfort standards in MATLAB. Finally, prediction model based on back propagation (BP) artificial neural network (ANN) was constructed, and genetic algorithm (GA) was adopted to optimize the connecting weights and thresholds of BP ANN. Then, the prediction model based on traditional BP ANN and GA-BP ANN were respectively conducted as a case study to validate the feasibility and reliability of the method proposed in this paper. The results show that GA-BP prediction model has higher prediction accuracy and stronger generalization ability than traditional BP prediction model in prediction of ride comfort values.