Prediction of Railway Passenger Ticket Booking Quantity Based on Ensembles of Multi-step LSTM

Through analyzing the historical ticket booking quantity of railway passenger transport, this paper proposed a prediction method of railway passenger ticket booking quantity based on ensembles of multi-step LSTM. Firstly, several LSTM models were constructed as the first layer prediction model, and the Bayesian optimization method was used to train those LSTM models to obtain the candidate super parameter set. Secondly, the optimal parameters of those LSTM models can be obtained by substituting the candidate optimal super parameters into those LSTM models for the validation set. The weighted sum method, linear regression and lightGBM were used as the second layer models to fuse the prediction results of those LSTM models separately. By comparing and analyzing the prediction results of the above three methods in the second layer, it is found that the weighted sum method with the simplest calculation process can significantly improve the prediction effect, while the other two methods with slightly complex calculation can only slightly improve the prediction effect. The prediction method proposed is simple, robust and accurate, and can support the operation staff in the preparation of the emergency plan during the peak period.