Research On Short-term Forecast And Multi-path Assignment Of OD Passenger Flow In Urban Rail Transit

China’s strength in transportation has gradually turned to high-quality development.At the same time,the scale of urban rail transit network continues to expand,and passenger travel demand become more complex and diverse.Under the dual influence of policy direction and development demand,how to respond accurately and real-time to the change of passenger demand in the network has been widely concerned by the academic circle.The deployment of a large number of information systems provides a data basis for research.However,it is impossible to grasp the destination and route information of passengers.Therefore,how to solve the problem of short-term forecast and multi-path assignment of OD passenger flow is the core problem to be solved in this paper.On the basis of maximizing the use of AFC data,this paper uses deep learning,probabilistic reasoning and other technical means to establish a two-stage research framework of OD passenger flow forecasting and time-varying route assignment.The main work includes:(1)Establish the research foundation of short-term forecast of passenger flow and multi-path assignment.Sort out the factors that affect the distribution of OD passenger flow and passenger route choice and study the acquisition and processing methods of multi-source data(including the establishment of road network topology,extraction of passenger flow time series data,classification of network station types,and processing of historical weather data),which laid the data foundation for the following research;(2)Propose the "observation-estimate-forecast" short-time forecast model of OD passenger flow.Build the spatial and temporal correlation between OD flow and inbound passenger flow,and perform inference estimation from the multi-time domain historical data.Therefore,the estimated value of OD is obtained to make up for the missing problem of the pre-sequence OD data at the current moment.Furthermore,passenger flow characteristics,time characteristics,spatial characteristics and external characteristics are used as inputs to construct a multi-factor adaptive Kalman Filter-LSTM OD passenger flow short-term forecast model,which is trained for different forecast model scenarios to improve forecast accuracy;(3)Design the multi-path passenger flow assignment algorithm of "group resultsindividual behavior".The effective path set of OD pair is constructed,and the individual choice behavior is inferred from the group choice results presented by the data of swiping card combined with the travel time of passengers.The Gaussian mixture model of travel time of multi-path OD was solved based on EM algorithm,and the route choice preference of group passengers was obtained.The individual choice behavior is restored from the massive historical data,and the time-varying passenger route choice proportion coupled with the forecast granularity is calculated to improve the assignment accuracy.(4)Taking Suzhou rail transit network in 2019 as an example,short-time forecast is made for five pairs of OD,and the forecast results are analyzed from different passenger flow demands,operation days and operation periods;At the same time,multi-path passenger flow assignment was carried out for two pairs of OD,and the assignment results were analyzed from the group results and ticket type characteristics to verify the feasibility and applicability of the model.This paper starts from the two major links involved in the passenger travel chain,uniformly divides the granularity of forecast and assignment,makes up for the lack of quasi-real-time OD,improves the accuracy of short-term forecast of OD passenger flow,and restores individual passenger choice behavior.As a result,the operation and management capabilities of urban rail transit are improved,and information support is provided for online decision-making by the operation department.There are 49 pictures,37 tables and 76 references.