站城融合背景下高速铁路综合枢纽短时客流预测研究

Prediction of Short-term Passenger Flow of High-speed Railway Integrated Passenger Hub under Station-city Integration

站城融合背景下,短时客流预测是高铁综合枢纽运营组织、协同管理、应急反应的重要依据。通过分析客流相互影响机理,提出高速铁路综合枢纽铁路系统、周边土地利用与城轨系统间的客流拓扑关系;以枢纽站周边各类兴趣点数量表征土地利用特征,采用岭回归方法推算高速铁路综合枢纽周边土地吸引客流;基于铁路进站客流、枢纽周边土地吸引客流与城轨出站客流间的相互作用关系,提出基于图注意力网络的客流预测模型,同时引入多头注意力机制,自适应获取不同日期不同时段下铁路进站客流和土地吸引客流的权重,实现高速铁路综合枢纽城轨出站客流的精确预测。为验证模型有效性,以北京南高速铁路综合枢纽为例进行分析,结果表明该模型与SVR、LSTM和GCN等预测模型相比,其预测精度显著提升。

Under the background of station-city integration,short-term passenger flow prediction is an important basis for the operation organization,coordinated management,and emergency response of the high-speed rail integrated hub.Based on the analysis of the interaction mechanism between the subway outbound passenger flow in the high-speed rail integrated hub,this paper studied the passenger flow topological relationship between the high-speed rail station,the surrounding land use,and the urban rail system.The land use characteristics were characterized by the number of various points of interest(POI)around the hub station,and the ridge regression method was used to calculate the passenger flow attracted by the land use around the hub station.Based on the interaction between the inbound passenger flow in the railway station,the passenger flow attracted by the land use around the hub,and the outbound passenger flow of the urban rail station,a passenger flow prediction model based on the Graph Attention Network(GAT)was proposed.On this basis,a multi-head attention(MHA)mechanism was introduced to adaptively obtain the weight of railway inbound passenger flow and the passenger flow attracted by the land use under different time periods and different dates to accurately predict the passenger flow of the subway out of the high-speed rail integrated hub.In order to verify the validity of the model,an analysis was carried out with the Beijing South High-Speed Railway Integrated Hub as an example.The results show that the prediction accuracy of this model is significantly improved compared with other prediction models such as SVR,LSTM,and GCN.