Prediction Method Of Urban Rail Transit Dynamic O-D Based On Spatial-Temporal Characteristics Of Passenger Flow

With the rapid development of mass and network of Urban Rail Transit,the construction of an operation management and control system based on dynamic O-D(Origin-Destination)passenger flow will further promote the transformation of Urban Rail Transit Operation Management to Dynamic and intelligent direction.The prediction of Urban Rail Transit Dynamic O-D passenger flow can not only provide reference for operating organizations to make dynamic operation strategy,but also play an active role in passenger travel demand.To this end,this paper focuses on the research of the dynamic O-D prediction method of Urban Rail Transit,the main work is as follows:Firstly,an AFC based method is proposed to extract the spatio-temporal characteristics of Urban Rail Transit O-D passenger flow.Based on the pretreatment of AFC data,the spatial and temporal characteristics of O-D passenger flow are extracted from time and space dimensions by using various clustering analysis methods.By analyzing the stability and correlation of O-D passenger flow distribution,the credibility of the spatial-temporal characteristics of historical prior O-D passenger flow is calibrated.The passenger diversion rate and passenger contribution rate are introduced as the quantitative indexes of O-D passenger flow spatial-temporal characteristics,and the quantitative expression method of O-D passenger flow spatial-temporal characteristics is given.Secondly,the arrival rule of O-D passenger flow is analyzed based on travel time distribution,and an approach to calculate the arrival rate of O-D passenger flow based on travel time distribution is proposed.Based on the analysis of passenger travel time structure,this paper draws the histogram of travel time frequency distribution,excavates the dispersion of travel time,and extracts the rule of O-D travel time distribution based on K-S distribution test.In order to describe the arrival law of passenger flow,the arrival rate of passenger flow is introduced,and the calculation method of arrival rate of passenger flow based on travel time distribution is established.On this basis,combined with practical application,a more simple passenger arrival rate calculation method is proposed.Thirdly,a dynamic O-D prediction method based on improved Long Term Memory Network(LSTM)model is proposed.By analyzing the correlation of Passenger Flow Index,the problem of dynamic O-D forecast is transformed into the problem of O-D passenger flow spatial-temporal characteristics forecast.On this basis,the standard LSTM network is improved into a dual-input and dual-output model,and the corresponding training channels are set up for the time feature sequence and the spatial feature sequence of O-D passenger flow respectively,thus a dynamic O-D prediction model based on the improved LSTM network is constructed.The training process and weight matrix optimization algorithm of LSTM network are given.Finally,based on the AFC data of Nanjing Rail Transit,the validity of the dynamic O-D model proposed in this paper is verified.The training set data is constructed to train the model,the prediction effect of the model in different operation days,different operation periods and different types of O-D pairs is evaluated.By comparing with ARIMA model,Kalman filter model and standard LSTM model,the overall mean absolute percentage error(MAPE)of the dynamic O-D prediction model based on the improved LSTM network is 20.23%,7.5% and 3.17% lower than the three traditional prediction models,respectively,it shows that the model has better prediction effect.