Analysis Of Passengers’ Route Choice Behavior Of Urban Rail Transit Based On AFC Data

The network operation of urban rail transit system provides a variety of alternative routes for passengers to travel.However,the uncertainty of passengers’ route choice behavior brings challenges to train operation planning,new line access evaluation,ticket clearance and line network optimization of urban rail transit system.Therefore,the analysis of passenger route choice behavior is the basis of the above work.At present,there are two methods to determine passenger route choice: utility analysis method and data processing and mining method.The former can reveal the mechanism of passengers’ route choice behavior,but requires a large number of route choice preference data to support model calibration.The latter is based on historical Auto Fare Collection(AFC)data to calculate route selection results with high accuracy.Based on the above methods’ functions and data requirement characteristics,this study mainly carried out the following two parts of work: first,the dissertation improved the data processing and mining method,proposed the travel time pruning method,and transformed AFC data into “RP-like data”;Second,considering passengers’ semi-compensation behavior and heterogeneity,a route choice behavior model was constructed,and the mechanism of passenger route choice behavior was analyzed based on “RP-like data”.Specifically,for the data mining method,the dissertation analyzed the composition of passenger travel time first,and obtained the value of each part of passenger travel time through AFC passenger travel data and train diagram.Then,the shortest route and effective route of each OD pair were calculated respectively by Dijkstra algorithm and improved Depth-FirstSearch.Afterwards,on the basis of considering the calculation efficiency and data requirements,the dissertation chose Gaussian Mixture Model which has the characteristics of high calculation efficiency and less demand for data categories to establish model,and proposed the travel time pruning method to solve the problem of low clustering accuracy caused by similar travel time.Later,the EM algorithm was used to solve the model and the validity of the model was verified.Besides,the clustering results were " RP-like data".For the utility analysis method,starting from the establishment of the utility function,the dissertation determined the ride time,transfer time,transfer times,road network familiarity and comfort as the influencing factors,and defined the corresponding value ranges.Considering that there are two processes when passengers choose the route,namely passengers’ semi-compensation behavior,which are non-compensatory to choose the effective route and compensatory to determine the travel route,the route tolerance parameter was introduced into the utility function,and the CMNL model was established.Similarly,considering the heterogeneity among passengers,in order to facilitate comparative analysis,the Latent Class Model was selected to reflect the differences between passenger groups.Finally,the LC-CMNL model was constructed,which was verified by Z test and goodness of fit test.Subsequently,the dissertation took Chengdu metro as an example for analysis.After introducing the example background,the two route choice models were analyzed first.Among them,the clustering results of the Gaussian Mixture Model were consistent with the actual situation basically,and the significance level of each variable of LC-CMNL model was 95%or above.Besides,the LC-CMNL model fitting effect was superior to MNL model.Thus,the validity of each model was proved.Then,the dissertation analyzed passengers’ route choice behavior from three aspects.First,in terms of travel distance,the marginal utility of transfer times increased with the increase of travel distance,while the marginal utility of other variables decreased.Second,in terms of travel time,the sensitivity of passengers to travel time had little change,but the sensitivity to transfer time and transfer times were obviously higher in the peak hours of weekdays,while the road network familiarity and comfort were small.Last,as for passenger category,one-way card passengers had the greatest sensitivity to transfer times,and the sensitivity of regularity class 1 passengers to transfer times varies greatly in different travel times.Regularity class 3 passengers had the greatest sensitivity to the familiarity of the road network.However,the factors of ride time,transfer time and comfort were less affected by passenger types.