Recognition And Regularity Of Travel Behavior Patterns In Road Passenger Transport Driven By Ticket Sales Data

The new urbanization has reshaped the urban space and profoundly affected the daily trips of urban inhabitants.Meanwhile,for rural ones,immigration is accelerated likewise with its range extended and its process complicated,which leads to new changes of behavior patterns during inter-city travels.Over the past decades,road passenger transport has undertaken the great task of providing services for the travel demands derived from the new urbanization,on the strength of its flexibility and economy.However,with the rapid development of high-speed rail and private cars,it is suffering multiple impacts.In order to build a healthy and efficient comprehensive national transport network,it is required to explore high-quality transformation and development of road passenger transport,and the indispensable approach is based on the deep research on the travel behavior patterns of passengers.Based on the ticket sales data of road passenger transport,this dissertation studies the behavior patterns during inter-city travels,including the recognition model and the regularity analysis of travel characteristics,trip semantic patterns and time interval patterns of trips,from both perspectives of the individual and the group.Firstly,the methodology and framework of existing researches involving trip purpose identification,travel pattern clustering and regualarity analysis are reviewed and summarized,subsequently,the origin of the main ideas and the overall position of this study in the framework are interpreted.Secondly,based on the reconstruction of ticket sales data,the trip characteristics of individual and group passengers during inter-city travels are revealed respectively.On this basis,two enhanced probabilistic graph models(PGM)are proposed to accommodate and solve the problems of trip semantic patterns identification for group passengers and the time interval patterns clustering of trips for frequent individual passengers respectively.Finally,with the outputs of pattern recognition in the pervious step transformed into inputs,this dissertation further conducts quantitative analysis on the evolution of the ridership considering trip semantic patterns and the variability of behaviors considering time interval patterns of trips by adopting econometric models such as panel regression and multiple linear regression,which provides a theoretical support for practical work,such as optimization of the comprehensive national transport network.The major achievements attained in this dissertation are as follows:1.Taking road passenger transport as an example,this dissertation aims at the data cleaning and reconstruction algorithm for real-name ticket sales data.On this basis,it discovers that,from the perspective of individual passengers,1)the ridership has six typical periods in time and obeys the power law in space,and meanwhile,there are differences in the proportion of returning home in each period;2)and both the total number of travel times and destinations obey the power law,whereas the time interval of trips obeys the Poisson distribution associated with weekly and annual fluctuations;and from the perspective of group passengers,1)that the phenomenon that passengers tend to travel with companions in road passenger transport is sighnificant,meanwhile,the proportion of this pattern varies in time and space;2)and both of the number of the group members and potential companions obey the power law.2.Focusing on group passengers,this dissertation solves the problem of trip purpose completion for ticket sales data without ground truth but with low spatial resolution,meanwhile,trip purpose is extended as trip semantic patterns with consideration of its ambiguity.On the one hand,from the perspective of group passengers,features of companions and their relationships are also exploited to make better use of ticket sales data.Specifically,by viewing the passenger group,features of group members and trip purpose of a group as analogous to document,words in a document and topics of a document used in natural language processing,the problem of trip semantic patterns recognition is defined as topic mining,and a Time Topic Model(TTM)is built to address this problem.On the other hand,the features are extracted with consideration of the unique nature of ticket sales data,including the demographic,experience and co-travel network features,furthermore,the discretization and textualization methods for feature processing are put forward.Through additional travel surveys,the efficiency of the proposed features and TTM are evaluated,and a more accurate,stable and balanced result is obtained compared with the benchmark models.Finally,a case study is conducted and accordingly,four primary trip purposes(i.e.official business,returning home,journey and personal business),as well as unconventional ones beyond exsiting knowledge domains,are clustered and labelled based on the distributions of features and start time,which are estimated by Gibbs sampling algorithm.The completion of trip purpose is useful to enhance the depth of ticket sales data in travel demand modeling.3.Focused on frequent individual passengers,this dissertation reveals the temporal pattern and the regularity of the behavior during inter-city travels.To bridge the gap encountered that the time of different year cannot be aligned due to the combination of solar and lunar calendar,start time is replaced by the time interval,in which case the frequency can be reserved and the time scale can be unified.Moreover,the time interval is calculated not only under the absolute time coordinate but also under a relative time coordinate,and through distinct measurements,the existence of intrinsic formation mechanisms of time interval patterns is confirmed.To address the problem of clustering time interval patterns of trips at the microscope level,a bi-level Gaussian Mixture Model(BLGMM)is built using the representation of bag-of-words.Correspondingly,a two-step parameter estimation method combined with an initialization step by preliminary clustering and a fine-tuning step by an extended Expectation Maximization(EM)algorithm is provided.Furthermore,by cross-correlation analysis of the results among age,gender,the departure date,the departure period and number of companions,it is concluded that there are significant differences in the characteristic distributions of different patterns.These findings are helpful to understand the temporal pattern and the regularity of the behavior during inter-city travels at a microscopic level,to support personalized and demand-responsive travel services,and in order to promote the transformation from incremental passengers to inventory passengers.4.On the basis of the results obtained from the pattern recognition by PGM,regularity of behavior patterns during inter-city travels is studied by using econometric analysis.In terms of the evolution mechanism of annual riderships,panel data about the riderships of different districts are collected at first,with a further decomposition of the total based on trip purposes,and with a further division of districts depending on the opening of high-speed rail;second,18 panel regression models are then introduced to explore the factors and the hysteresis effects at different scenarios;at last,the model estimation and the robustness test indicates that the increase of secondary road significantly raises the ridership of returing home and journey.As for the generation mechanism of variability in time interval patterns of trips,indicators that measure the local standard deviation and the local coefficient of variation in conditions of a mixture model are proposed at first;multiple linear regression models are then introduced to explore the mechanism of the variability.