Research On Commuters’ Travel Identification Method And Temporal And Spatial Characteristics Of Urban Rail Transit Passengers

With the extensive operation of urban rail transit,a large amount of passenger travel data can be accumulated.Studying the laws in this data has changed the shape of the city at the micro and macro levels.At the same time,it also promotes the optimization of passenger service management to reduce labor costs and improve transportation efficiency.The use of traffic smart card data to better understand the commuter behavior and mine its temporal and spatial characteristics has become a research hotspot.At present,the research on the commuter passenger flow law of urban rail transit mainly focuses on the exploration of the existing passenger flow law.Taking each passenger as the research unit,the commuting mode is obtained by studying the travel behavior of each passenger in a long time span.Although there are various methods to identify the activity mode of commuter passenger flow and its place of work and residence,and many analyses have been made based on the identification results,few studies have specifically analyzed the accuracy,reliability and sensitivity of the identification method itself,so that all the analysis results based on the existing identification methods are not based on a solid foundation.In this paper,the application effects of three representative methods are comprehensively compared by using the traffic data,and based on the comparison results,a new job and home recognition model is proposed to deeply explore the temporal and spatial characteristics of commuting patterns.This paper mainly completes the following work:Firstly,based on the swiping card data of rail transit in Beijing for five consecutive weeks in 2016,three representative commuting methods based on the staying time,trip number and hidden Markov model are applied,and the number of commuters at each station and the distribution of jobs and home in the whole network are deeply studied;The temporal and spatial differences of the three recognition results are discussed from the perspective of aggregated,the recognition characteristics between different methods are studied from the quantitative cross relationship,and the stability and operation efficiency of the three methods are quantified from the evaluation index.In conclusion,the three methods are summarized from five aspects: method characteristics,data object,calculation difficulty,advantages and disadvantages.Secondly,based on the above discussion results,a new job and home recognition method based on K-means+HDP model is proposed.The passenger trip chain is constructed by using the swiping card data,from which the passenger travel law is analyzed,and three time indexes of activity duration,activity date and activity arrival time are extracted.Based on Baidu map,the POI data within 500 meters around rail transit stations are statistically analyzed,and four site indicators of different service types are clustered by K-means algorithm.Combined with the four indexes as input parameters,a hybrid model based on hierarchical Dirichlet process is constructed to find four interpretable travel modes from all activity records.Thirdly,the HDP model and the other three models mentioned above are comprehensively compared by using the Perplexity evaluation index,the analysis of cluster activity characteristics and the cross relationship.Finally,the database is constructed based on the commuter travel records identified by HDP model.And seven clustering indexes are selected from regularity,timeliness and space by using the importance method of predictive variables.The commuter travel is divided into two representative modes by using two-step clustering algorithm,and the two types of modes are deeply discussed from the temporal and spatial characteristics.