| "The Fourteenth Five-Year Plan for the National Economic and Social Development of the People’s Republic of China and the Outline of Long-term Goals for 2035" proposes to accelerate the development of the total scale of urban rail transit,which effectively support national strategies such as new urbanization,metropolitan and regional integrated development.In the future,a large number of new rail transit lines will be planned,constructed and operated in China.The passenger flow forecast of urban rail transit is an important basis for the feasibility analysis of the construction of the new line,and the operating unit to formulate an operation management plan.At present,the four-stage method driven by the model is mainly used to predict the passenger flow of new rail transit lines.This method is subjective and the prediction effect is not ideal.Therefore,from a data-driven perspective,this thesis argues an accurate and efficient method for predicting passenger flow of new rail transit lines.First of all,this thesis analyzes the characteristics of rail transit passenger flow and its influencing factors,and elaborates the whole process of passenger travel on the metro from the three aspects of passenger flow generation,transfer,and connection.It provides an explicit description of the passenger flow characteristics from two aspects: the size and direction of passenger flow at rail transit stations and the average station riding distance.On this basis,it analyzes the influencing factors of rail transit passenger flow from the four aspects of regional socio-economic indicators,land use,metro network construction level,and station connection level.Moreover,it explains the key issues in the forecast of passenger flow in terms of data and models.Secondly,this article studies the multi-dimensional data of rail transit,which lays the data foundation for the passenger flow prediction of the new line access.Multi-dimensional urban rail transit data includes two parts: rail transit AFC transaction data and rail transit station external data.The Automatic Fare Collection transaction data of rail transit are preprocessed and analyzed statistically,and the characteristics of station passenger flow are counted.Besides,the determination of the passenger flow attracting area of rail transit stations is the prerequisite for studying the external data of rail transit stations.Representative stations are selected to conduct a survey on walk and bus connection time,then the connection time threshold is determined.Furthermore,transit station catchment areas are determined based on the time threshold.The factors affecting rail transit passenger flow are quantified within the station catchment areas,and the external data acquisition and calculation methods of rail transit stations are given.Finally,the data-driven passenger flow prediction of new lines is researched,including the station passenger flow forecast and the distribution forecast.A station passenger flow prediction model based on Radial Basis Function neural network is constructed.The external data of stations in existing years are used as the model inputs,and the station passenger flow characteristics of the corresponding year are exploited to the model outputs for neural network training.The inbound and outbound passenger flow and the average riding distance of the station is predicted through inputting the external data of stations into the RBF neural network model.In addition,comprehensive factors that can characterize the attributes and location of the station are obtained by factor analysis method which applied to construe external data and average riding distance of stations.The comprehensive factors are input to the hierarchical clustering model,and the rail transit stations are divided into nine categories.Based on the station classification results,the distribution of passenger flow between different types of stations are studied by improved gravity model.The least square method is used to calibrate gravity model parameters by using the existing network OD passenger flow data.The passenger flow distribution is predicted later.An example indicates the method which is driven by multidimensional data is more accurate than the traditional four-stage prediction method. |
