The prediction of railway vehicle vibration based on neural network

Vehicle vibration acceleration is an important parameter reflecting the state of track irregularities and the quality of wheel and rail contact. A model of predicting vehicle system based on Nonlinear Autoassociative time-series network with external input (NARX) Neural Network (NN) is established by using the vehicle dynamics model, which relates track irregularities to vehicle performance. To improve the prediction precision, the properties of the NN were investigated, and, thus, the number of time delay and hidden nodes were determined. The validation results of the simulation data from SIMPACK model show that the prediction output of NARXNN is highly relevant with target output with smaller Root Mean Square (RMS) error. Meanwhile, the model can well predict the vehicle vibration acceleration with a consistent result. The model was validated based on measured data, further proving that the proposed prediction model of NARX NN is able to predict the output of vehicle vibration acceleration with high precision. © 2015 by Binary Information Press