Spatial-Temporal Evolution Characteristics And Prediction Method Of Passenger Flow At Rail Transit Stations Under Large-Scale Events

As the practitioner of green and low-carbon transportation and the main artery of urban public transportation,the urban rail transit has become an important way to alleviate the contradiction between transportation supply and demand due to its advantages of strong accessibility,high efficiency,and sustainability.In recent years,with the continuous enrichment of people’s cultural life,various cultural and sports events have been held frequently.In special scenarios of large-scale events,the passenger flow of rail transit stations breaks the conventional passenger flow demand pattern,exhibiting abruptness and aggregation,which will affect the safe and efficient operation of the rail transit system.To reasonably guide operation management,effectively prevent emergencies,and scientifically guide passengers to travel,it is necessary to research the spatiotemporal evolution characteristics of passenger flow at rail transit stations,the prediction of inbound and outbound passenger flow,and the detection and warning of sudden passenger flow in large-scale event scenarios.Existing studies mostly focus on conventional passenger flow prediction,while little research has focused on the characterization of the spatiotemporal evolution characteristics of passenger flow in special scenarios such as large-scale events,and the prediction and emergency management of station passenger flow.This paper analyzes the spatiotemporal distribution and evolution characteristics of urban rail transit station passenger flow,constructs a combined optimization prediction model of inbound and outbound passenger flow based on multi-dimensional features,and proposes a large-event-oriented rail transit inbound and outbound passenger flow warning and control method,so as to achieve accurate prediction of station passenger flow under large-scale events and effective detection and early warning of abnormal activity passenger flow.Specifically,the main contributions are summarized as follows:(1)The spatiotemporal distribution and evolution characteristics of passenger flow at rail transit stations are analyzed in detail.Firstly,the data sources and processing methods are introduced,and the influencing factors of the station passenger flow are analyzed.Then,the temporal and spatial distribution characteristics of conventional passenger flow are studied by the mathematical-statistical method.Finally,the spatiotemporal evolution law of passenger flow at rail transit stations is discussed with the information on large-scale events.(2)A multi-dimensional feature extraction method is designed for the passenger flow of rail transit stations.Firstly,the agglomerative hierarchical clustering method is combined to determine the morning and evening peak hours of passenger flow,the grey relational analysis is used to identify spatially related stations with similar passenger flow demand patterns,and the passenger flow shock coefficient is introduced to determine the impact period of large-scale events.Then,three-dimensional features of time,space,and event are constructed based on station passenger flow data and large-scale event information.In addition,the numerical characteristics are filtered by the Elasticnet algorithm to solve the problem of selecting the optimal characteristic data set of passenger flow at rail transit stations.(3)A combined optimization prediction model is constructed for passenger flow at rail transit stations.Based on multi-dimensional feature input and ensemble algorithm theory,the Bayesian optimization principle and error weight combination strategy are introduced to construct a station passenger flow prediction model based on Light GBM-Cat Boost combination optimization.Then,an example is given to illustrate the prediction accuracy and robustness of the model in the prediction of inbound and outbound passenger flow in large-scale event scenarios.The result shows that the average prediction accuracy of the model in the inbound and outbound passenger flow prediction of the three stations is 93%.(4)A large-event-oriented passenger flow forecasting and control method for rail transit inbound and outbound passenger flow is proposed.The STL time series decomposition is introduced to eliminate the influence of large passenger flow in the morning and evening peaks,and the i Forest detection algorithm is used to diagnose the abnormal growth of outbound passenger flow before the large-scale event,so as to realize the purpose of identifying the occurrence of the large-scale event in advance and sending the event warning.Then,the suggestions for the control of sudden inbound passenger flow caused by the large-scale event are put forward from the aspects of passenger flow organization management and passenger flow classification control.