Research On Analysis Method And Model Of Urban Rail Travel Behavior Based On Multi-source Data

The rail travel behavior characteristics of urban residents are important data support for operations such as urban rail cross-sectional passenger flow forecasting,formulation of line network planning schemes and station TOD development analysis.The complex rail network distribution and growing passenger volume in China’s cities bring challenges to the subsequent rail network planning,traffic organization and station passenger flow control.To build a high-quality urban rail transit operation system,it is necessary to grasp the accurate characteristics of passengers’ rail transit travel behavior,understand the real-time direction trend of passenger flow within rail transit,clarify the characteristics of interchange station passenger flow distribution,and analyze the mechanism of urban rail passenger transportation choice behavior.Conventional single data methods are insufficient in its ability to collect the complete travel chain information of rail transportation,and there are often errors or omissions.Traditional transportation choice behavior prediction is mainly based on questionnaire data,where the accuracy of the model only relies on the fit and hit rate assessment,in lack of real data verification.Meanwhile,they are conducted without detailed analysis of travel crowd characteristics,leading to relatively rough selection of random parameters of the model utility function,and therefore needs further refinement.Due to difficulties in data acquisition,incompleteness of data information,inconsistency between data and model structures,and how to integrate and utilize data from different sources,the current research on rail passenger flow characteristics and supporting rail travel behavior modelling using multiple sources of data such as mobile phone signalling,rail AFC,and travel behavior surveys are still in the exploratory stage,and requires further research.Therefore,this thesis uses multi-source data such as mobile phone signaling,rail AFC,travel logs,GPS and travel behavior surveys to study travel behaviors including travel paths inside the rail transit and the OD outside the stations.On this basis,a mixed logit model for rail behavior selection is constructed and optimized,and a behavior model parameter optimization that fuses multi-source data,where the optimal model parameters are analyzed based on crowd classification.The main research results are as follows.1.The correlation between multi-source data and rail travel behavior is investigated.This thesis investigates the characteristics of each of the four types of data(mobile phone signaling,rail AFC,travel logs and GPS and travel behavior surveys),the travel characteristics that can be obtained and their correlation with travel choice behaviors.It explores how to better utilize the complementarity between multiple sources of data to identify rail travel behavior features.2.The collection and processing methods of multi-source data are studied,and schemes for travel log survey experiments and travel questionnaire surveys are designed to provide basic data support for rail travel behavior research and choice behavior model construction.The thesis discusses the time,scopes and methods of collection of different data,and takes Chongqing city as an example to collect multi-source data such as mobile phone signaling data and rail AFC data simultaneously,and proposes processing methods for various data,as well as enrolling12 volunteers to carry out a 30-day travel log data collection experiment,in which a total of 1000 questionnaires are collected online and offline.3.A method for the identification of internal travel paths based on K-short-circuit and Niedermann-Wunsch algorithm is proposed.To address the problem that the traditional extraction of internal paths is mostly based on the theoretical model projection of aggregate samples of AFC data,and the accuracy of individual travel feature identification needs to be improved,the thesis proposes an individual intra-track travel path identification algorithm fusing mobile phone signaling,AFC and travel log data.In the calibration of the rail base station library,the volunteers integrated the DBSCAN density clustering algorithm to calibrate the base station library of the rail stations and lines,as well as the above-ground and underground stations through multiple trips.In the process of identifying rail entry and exit stations,based on the sequence of base stations connected by passengers to identify the inbound and outbound behavior of rails,considering the adjacent relationship between stations,a K-short-circuit algorithm is proposed to identify individual passenger travel and transfer stations.It adopts the Niedermann-Wunsch algorithm to match the similarity of K-bar base station sequences with the those of passenger internal travel connections to avoid the influence of subjective settings of traditional methods,so that complete individual rail transit travel path information is constructed and compared with the volunteers’ travel log and AFC data,respectively.The accuracy of volunteers’ signalling data in and out of stations is 81.93% and 80.58% respectively,and the accuracy of interchange station identification is 80.33%.The accuracy of individual entering and exiting stations is 82.23% and 80.25% respectively.The proposed method based on K-short-circuit and Niedermann-Wunsch algorithm for rail transit internal travel path identification can effectively recognize individual intra-track travel information,and meet the needs of rail transit operation management.4.A new method of OD location point identification outside rail transit stations based on multi-source data and spatio-temporal density clustering algorithm is constructed.For the clustering algorithm of travel OD location point identification based on mobile phone signaling data,the temporal and spatial threshold parameters are subjectively set,and lack of optimization selection algorithm for clustering parameters,which leads to weak model universality and interference resistance.This thesis uses multiple sources of data,including mobile phone signaling,travel log and GPS data of the volunteers.Based on the spatio-temporal density clustering algorithm,parameters of two optimized clustering algorithms,fused genetic algorithm and simulated annealing algorithm,are used to construct a new method for the identification of OD location points outside individual rail travel stations.The empirical results show that the genetic algorithm results outperform the simulated annealing algorithm.By comparing the data such as actual volunteer travel logs and GPS data,it is known that the average error in the identification of OD location points of the volunteer travel outside the stations is 634 m.5.The optimization method of transportation choice behavior model based on multi-source data is established.Based on multi-source data in the forms of mobile phone signaling,rail AFC and travel behavior survey,the thesis proposes a behavioral model optimization method for rail sharing ratio calculation and crowd classification by analyzing the effects of variable selection,parameter estimation sampling method and heterogeneous parameter distribution on the prediction accuracy of the mixed Logit model.On the one hand,a mixed logit model for rail behavior selection is constructed,adopting a stepwise regression method to optimize the selection process of the explanatory variables of the travel behavior model,and integrating the results of the rail share ratio calculation with mobile phone data to quantify the model prediction process and effects of different behavior model base theories,modelling methods and random parameter distributions,sampling methods,etc.,thus a behavior model optimization method is proposed to improve the prediction accuracy of rail share ratio.On the other hand,the optimization method of the behavioral model parameters by fusing multiple sources of data is proposed,where the optimal model parameters are analyzed based on crowd classification.The results of the study show that the random parameter distribution and the selection of sampling method have a large impact on the prediction accuracy of the model,and the model has the best predicative validity when the normal distribution random parameters and variable order Halton sampling method are used,with a prediction error less than 6.38%.The prediction error of the behavior model selected by optimizing the parameters based on crowd classification is only3.82%,and the prediction accuracy is improved by 2.56%,indicating the feasibility and effectiveness of the travel choice behavioral optimization model using multiple sources of data.