Research On Ridership Prediction Method Of Urban Rail Transits Stations Based On Multi-Source Data

With the rapid development of rail transit of China,some urban rail transit networks have begun to take shape,showing a large ridership.Due to the geographical location of rail transit stations,the surrounding built environment and its own functions and other factors,the spatial and temporal distribution of ridership at the stations is unbalanced.Therefore,through multisource data such as AFC data,POI data and road network vector data,this manuscript analyzes the correlation between the station ridership and the built environment around the station,and puts forward a prediction method of station ridership considering spatial heterogeneity,which provides a theoretical basis for rational allocation,planning and design of facilities around the station.This study uses python to crawl the POI data around the rail transit station through the interface of Gaode Map Application Program,and uses all kinds of POI data and their numbers to represent the land use types and development intensity around the station.According to the number of bus stops around the rail transit station,its service lines,road network density and the number of stops arriving at other stations in different time periods,the station accessibility is reflected.Combined with the station’s own attributes,a set of influencing factors including the built environment around the station,station accessibility and station attributes is established.According to the obtained AFC data of rail transit,this manuscript analyzes the characteristics of station ridership in different periods of networking on weekdays,weekends and holidays from two dimensions of time and space.The stations are divided into six types by Kmeans clustering method,and the time distribution law of ridership in and out of different types of stations is analyzed,and the mechanism of the influence of the built environment around the stations on the ridership is revealed.Through the OD ridership data of each rail transit station,this manuscript analyzes the correlation of ridership among stations under the condition of network formation.According to the different contribution degree of ridership influencing factors of each rail transit station to ridership,a candidate variable set of ridership influencing factors is established,and the multiple linear regression model is used to reveal the relationship between ridership and ridership influencing variables of each station.Through spatial autocorrelation,multicollinearity and model significance,the key variables affecting ridership at rail transit stations are screened,and the least square regression model is constructed to predict ridership at stations.Considering the spatial heterogeneity of the influencing variables at stations,based on the geographic weighted regression model,a forecasting method of station ridership is proposed.Finally,taking Nanjing rail transit as the research object,the data of ridership and influencing variables of each rail transit station are extracted,and four key variables affecting ridership at stations are screened out,which are substituted into the constructed station ridership forecasting model for solution.Six types of different stations are selected to forecast,and the reliability of the model is verified.Through the average absolute error and root mean square error indicators,the results show that the forecast accuracy of ridership at stations based on the geographically weighted regression model is higher and the fitting effect is better.