Train delay prediction is a significant part of railway delay management, which is also the theoretical basis and the basis of timetable optimization. Thus, a shallow extreme learning machine model is proposed to effectively predict train delays. Specifically, speed limit, percentage of journey, train delays and stations influenced are taken as independent variables. More importantly, having the characteristic of no need to adjust the parameters of hidden layer, the training stage of this model is tremendously simplified. A case study of one busy domestic line is illustrated and the comparison with the random forest model is further done to demonstrate the accuracy of the proposed model. The results show that the proposed model can accurately predict the delay time at the next station the train will arrive from history data, with an accuracy rate of 0.8, outperforming the random forest model with an accuracy rate of 0.6.

CICTP