Prediction Of Short-Term Incoming Passenger Flow Of Subway Station Connecting The Railway Passenger Hub

In recent years,with the rapid development of China’s transportation construction and urban economy,railroad passenger hubs have become the object of vigorous development,as the pivot of passenger transportation connection and conversion inside and outside the city,integrating various transportation modes such as railroads,ground buses,urban railways and cabs,and being an important node of the comprehensive transportation system of large cities.Due to its better accessibility and affordability,urban rail transit has become an important way for passengers to transfer and connect at railroad passenger hubs.Compared with ordinary urban rail transit stations,urban rail transit stations connecting to railway passenger hubs have more complex passenger flow formation principles and change laws,and their passenger flow prediction methods are also very different from those of ordinary subway stations.In this paper,we take the inbound passenger flow of metro stations connected to railway passenger hubs as the research object,model the inbound passenger flow of railroad interchange metro stations from the perspective of passenger flow formation mechanism,consider the influence of railroad arrival passenger flow and complex spatial structure in the hub,and build a combined MHA-GCN-LSTM prediction model to achieve accurate prediction of inbound passenger flow of metro stations connected to railroad passenger hubs.The main work of this paper is as follows:(1)Analysis of the inbound passenger flow characteristics of the subway connected to railroad passenger hubs.Taking Beijing West Station,National Library Station(interchange station)and Liuliqiao East Station(general station)of Beijing Metro Line 7 as examples,we compare and analyze the daily changes of passenger flow characteristics of each station.The results show that,in terms of time distribution,unlike ordinary stations with obvious morning and evening peaks,the inbound passenger flow of subway stations connected to railroad passenger hubs is more influenced by the arrival of railroad trains and shows obvious impulsiveness;moreover,the inbound passenger flow of subway stations connected to railroad passenger hubs is significantly higher than that of other stations on the line,especially during peak hours,the inbound passenger flow is still at a high level,while the inbound passenger flow of ordinary stations The inbound passenger flow during peak hours is generally lower.On this basis,the prediction problem of the subway passenger flow at the connecting railroad passenger hub is analyzed from the formation mechanism of passenger flow,and the network structure of the inbound passenger flow graph is constructed,which lays the foundation for the modeling and prediction model of the railroad interchange subway passenger flow in the later paper.(2)The modeling of railway-to-subway passenger flow.In order to characterize the time distribution pattern at each node of the passenger flow network diagram structure,to characterize the guiding relationship between the railroad arrival passenger flow and the subway inbound passenger flow,to analyze the internal mechanism of the time distribution pattern of the arrival passenger flow in the railroad passenger hub and the mechanism of the interaction between the passengers and the external conditions such as station infrastructure,to construct mathematical analysis models for each stage of the process of the railroad arrival passenger transfer to the subway in the hub.A mathematical analysis model is constructed for each stage of the process of transferring passengers to the subway in the hub to characterize the temporal distribution of passenger flow.Taking Beijing West Station as an example,the temporal distribution of passenger flow is derived based on the simulation idea,and compared with the temporal distribution of passenger flow extracted from AFC data,the overall distribution trend is more consistent and the accuracy reaches more than 90%,which proves the validity of the proposed passenger flow model and provides effective passenger flow input for the subway inbound short-time passenger flow prediction model connected with the railroad passenger hub.(3)The MHA-GCN-LSTM metro short-time inbound passenger flow prediction model is constructed.To address the difficulties of predicting the incoming subway passenger flow at the connecting railroad passenger hub,we propose the MHA-GCN-LSTM based subway short-time incoming passenger flow prediction model to obtain the time dependency relationship between passenger flows at each node in the passenger flow network diagram structure,characterize the spatial structure of railroad arrival passengers transferring into the subway,and obtain the influence weights between node features at different dates and times through multi-headed attention calculation,adaptively.(3)The spatial structure characteristics of the nodes in the subway,and the influence weights of the nodes’ features at different times of the day are obtained adaptively through multi-headed attention calculation,so as to realize the accurate prediction of the incoming passenger flow of the subway connecting the railroad passenger hub.(4)Example analysis of Beijing West Station.Taking Beijing West Station as an example,the railroad arrival passenger flow,surrounding traffic source-generated passenger flow,railroad exit passenger flow,railroad transfer subway passenger flow,subway inbound passenger flow and the constructed passenger flow network diagram structure relationship of Beijing West Station for the previous two 15 min are used as inputs to predict the inbound passenger flow of Beijing West Station subway station for the next 15 min.The prediction results are compared with the baseline models such as ARIMA,SVR,LSTM and GCN,and the results show that the average absolute error,root mean square error and average percentage error of the MHA-GCN-LSTM model proposed in this paper are smaller than those of other baseline models,and the model has the shortest running time,and the residuals of the prediction results are generally smaller and more concentrated,showing the optimal prediction performance.The effectiveness of the MHA-GCN-LSTM model proposed in this paper is verified.