Abstract: The prediction of safety size of EMU wheel set provides a basis for EMU safety evaluation.Due to the complexity of wheel set size change affected by operating environment and other factors, this paper proposed a particle swarm optimization multi-kernel extreme learning machine (PSO-MK-ELM) prediction model which was suitable for wheel set size data.The multi-kernel (MK)function, which was composed of polynomial kernel function and radial basis kernel function, was introduced into the extreme learning machine, and four key parameters of the model were optimized by particle swarm optimization.According to the wheel diameter data of CRH2 model, the rationality and accuracy of this method were verified by comparing the prediction results of different algorithms.The prediction results show that PSO-MK-ELM model can obtain better goodness of fit, mean square deviation, mean absolute error and mean absolute percentage error than BP model, ELM model and three kinds of commonly used KELM model in the prediction of EMU wheelset size data, which verifies the validity of the model in the prediction of EMU wheelset size.