The number of passengers in a high-speed railway line normally varies significantly by the time periods, such as the peak and nonpeak hours. A reasonable classification of railway operation time intervals is essential for an adaptive adjustment of the train schedule. However, the passenger flow intervals are usually classified manually based on experience, which is subjective and inaccurate. Based on the time samples of actual passenger demand data for 365 days, this paper proposes an affinity propagation (AP) algorithm to automatically classify the passenger flow intervals. Specifically, the AP algorithm first merges time samples into different categories together with the passenger transmit volume of the stations, which are used as descriptive variables. Furthermore, clustering validity indexes, such as Calinski–Harabasz, Hartigan, and In-Group Proportion, are employed to examine the clustering results, and reasonable passenger flow intervals are finally obtained. A case study of the Zhengzhou-Xi’an high-speed railway indicates that our proposed AP algorithm has the best performance. Moreover, based on the passenger flow interval classification results obtained using the AP algorithm, the train operation plan fits the passenger demand better. As a result, the indexes of passenger demand satisfaction rate, average train occupancy rate, and passenger flow rate are improved by 7.6%, 16.7%, and 14.1%, respectively, in 2014. In 2015, the above three indicators are improved by 5.7%, 18.4%, and 14.4%, respectively.
A Novel Hybrid Model for Short-Term High-Speed Railway Passenger Demand Forecasting
