Research On Passenger Clustering And Passenger Flow Distribution Of Urban Rail Transit Based On AFC Data

With a large number of new lines put into operation,the urban rail transit operation line network has been constantly improved,with increasingly complex structure and increasingly obvious network operation characteristics.The complexity of rail transit network topology makes the route choice behavior of passengers on rail transit network more and more complex.At the same time,the increasing heterogeneity of passengers brings great challenges to the passenger flow of rail transit.Traditional passenger flow allocation relies on survey data,which is one-sided,arbitrary and costly,so it is difficult to ensure the dynamic adaptability of the algorithm.The continuous accumulation of big data of passenger flow in rail transit provides perfect stock data for passenger flow allocation based’ on passengers differences.Therefore,this study starts from the exploration of passengers’ travel time and space characteristics,classifies passengers from the perspective of travel rules based on long-term travel information of traffic CARDS,and conducts differentiated passenger flow distribution for different categories of passengers.Specifically,this study mainly completed the following aspects of work:First of all,passenger clustering is carried out according to passenger characteristics.By using AFC data of subway passengers with long time span to extract travel time and spatial characteristics,regular passenger screening is conducted from the aspects of travel frequency and spatial consistency.The topic model is applied to passenger classification from the perspective of difference of travel time characteristics.Ten travel topics were selected,and on this basis,passengers were divided into 6 categories by k-means clustering algorithm.By analyzing the travel time and spatial distribution rules of different categories of passengers,we can have a preliminary grasp of their social and economic attributes.Secondly,the multi-logit model is adopted to establish the passenger flow distribution model of urban rail transit,and the algorithm framework is established considering the influence of travel time transfer cost,road network familiarity,congestion and other factors.In terms of parameter calibration,the traditional questionnaire survey was abandoned.The parameter calibration method based on bayesian inference was used to solve the problem by combining MCMC algorithm with AFC data or travel time survey data.The parameter sets corresponding to each type of passenger travel characteristics were obtained.Finally,taking Beijing subway line network in 2016 as an example,the method above is used to allocate passenger flow of each category in turn,and the results of daily passenger flow are integrated.In order to analyze the passenger flow in the peak period,the virtual OD matrix was extracted to allocate the passenger flow.The results of cross-section passenger flow,passenger flow statistical index and transfer passenger flow during the whole day and the peak time are calculated,which verify the feasibility and applicability of the model.On the basis of peak passenger flow,the change of passenger route selection probability under the influence of congestion is analyzed,which shows that the congestion index has significant influence on the result of passenger flow allocation.The passenger classification and passenger flow allocation method based on AFC data can determine the classification results and calibrate the parameters according to the travel data in different time periods.The method has wide sample range,strong timeliness,and can be dynamically adjusted,and the results have quantitative basis.It improves the traditional algorithm of passenger flow distribution,strengthens the universality of the algorithm,and provides reference for the distribution of passenger flow under the network operation.