| With the increase of urban rail transit network scale and passenger flow,the adaptability between operation scheme and time-varying passenger flow is obviously insufficient,and the coupling control between transportation capacity and passenger flow demand needs to be strengthened.Therefore,it is necessary to improve the dynamic and intelligent operation management level based on information data mining.The key foundation is to master the passenger travel rules,the evolution of passenger demand and the real-time OD passenger demand.Therefore,based on AFC data and POI data,combined with statistical analysis,data mining,machine learning and other methods,this paper focuses on passenger travel mode and dynamic OD estimation.This paper mainly includes the following aspects:(1)Based on the AFC data of urban rail transit and the POI data of 800 meters radius around the station,the passenger travel mode and its relationship with land use types are analyzed.Based on the screening of passengers with larger rail transit travel intensity,the database of individual passenger travel mode is constructed.The statistical analysis and data mining of passenger travel modes are carried out.The OD travel ratio and OD travel intensity are innovatively proposed to describe the passenger spatial travel mode.The two-step clustering algorithm is used to mine the travel modes of multiple trips and single trips.Finally,five types of passengers including two types of commuters(single commuter and multiple commuter)are obtained,and the daily travel time characteristics of different types of passengers are analyzed.The spatial characteristics of commuter passengers’ travel are analyzed.Based on the spatial travel mode of commuter passengers,the identification method of job-housing locations of commuter passengers is proposed.The relationship between land use type and travel mode is analyzed by job-housing ratio.(2)Analyzing the relevant characteristics of passenger OD can provide support for dynamic OD estimation.Based on commuter passenger travel mode,a fixed passenger recognition rule is constructed,and its recognition accuracy is above 97%.The correlation coefficient is used to analyze the correlation of OD passenger flow in different operation days,and the cycle rule of OD passenger flow is quantified.The operation time for dynamic OD estimation is 6:00 to 23:00,and the minimum estimation period is 15 minutes.The relationship between inbound and outbound passenger flow and OD passenger flow is constructed.The ratio of standard deviation of travel time to average value is used to describe the stability of OD to travel time,and a method to calculate the arrival rate of passenger flow by historical travel time distribution is proposed.(3)According to the requirements of dynamic operation management,a dynamic OD estimation model is constructed.A dynamic OD estimation method based on gated recurrent neural network(GRU)combined with particle filter(PF)is proposed.The dynamic OD prediction model based on GRU is used as the state transition equation of the state space model of PF to construct the GRU-PF dynamic OD estimation model.The calculated results of the model are superimposed on the fixed passenger OD identified based on real-time AFC data,which obtains the dynamic OD estimation results.(4)Taking Beijing subway as an example,the reliability and applicability of the dynamic OD estimation method are verified.The conclusion analysis shows that considering the fixed passenger travel pattern can reduce the dynamic OD estimation error,based on which a dynamic OD estimation scheme with high time efficiency and high accuracy is proposed. |
PDF Full Text Request