Data-based Passenger Route Choice And Dynamic Estimation Of Passenger Flow Congestion In Urban Rail Transit Network

With the continuous access of new lines and the development of urban rail transit network,the methods to establish a more accurate route choice model and predict the dynamic changes of passenger flow of urban rail transit network in a quick mode has become the preoccupation.Driven by real data,this paper focuses on the method and model of route selection and dynamic estimation of passenger congestion by means of machine learming.With the support of the dynamic deduction tool of passenger flow distribution,the dynamic change of passenger flow distribution in road network is formed,and the passenger flow congestion state in different situations is predicted rapidly.The main contents of the research include:(1)General modeling methods of passenger route choice and passenger flow dynamics in xirban rail transit network are analyzed.Using the data from passenger route choice results and passenger flow distribution simulation,the idea and research framework of employing machine learning to build a data-driven route choice model and passenger flow congestion dynamic estimation model are proposed.(2)Difference of route choice rules among different types of OD are analyzed using path selection data,for route choice considering heterogeneity of passenger flow is proposed.An ensemble learning framework and detailed steps of path selection based on fuzzy clustering of stations are given.Using FCM clustering algorithm to cluster Metro stations,the OD categories and their catagory membership degrees under different passenger flow composition are obtained,which provides a basis for training and combining path selection sub-leamers.(3)The SVR-based route choice sub-leamer is constructed,and the model training is carried out by using the data of passenger route choice.Furthermore,based on the fuzzy clustering results,an ensemble learning that combines route choice sub-learners is used to form a route choice decision model with higher accuracy.The prediction accuracy before and after the ensemble learming are compared and validated(4)Using the urban rail transit network passenger flow distribution simulation system as a tool for passenger flow derivation,a method of using machine simulation data from different simulation scenarios to estimate the congestion state of passenger flow is proposed.The LSTM-based passenger flow congestion dynamic estimation model in urban rail transit network is constructed and trained by simulating data.(5)A case study of Beijing urban rail transit after new lines was opened is carried out to verify the research method of this paper.