Rail Transit Station Ridership Forecast Based On Multi-level Influence Areas

Many large cities rely on Mass Rapid Transit(MRT)to increase passenger mobility.Compared with traditional zone-based travel demand models,direct ridership models(DRMs)that reveal associations of built environments and transit attributes with transit ridership at the station level are superior at estimating the benefits of transit-oriented development policies.The delimitation of catchment area around railway stations is the foundation of the direct risdership model.The catchment area always delimited by Arc GIS via a buffer analysis,which generates a non-exclusive or exclusive circular area.Studies have considered the road net and delineated the pedestrian catchment area based on network distance that are not be crossable by travelers are ignored in the process of generating catchment area.In this paper,considering the influence of barriers and multimodal transfer,the catchment is delimited by the cost distance tool in the Arc GIS.Firstly,three main transfer modes of rail obtained according to the transferring characteristics of rail transit transit,i.e.,walking,cycling and bus.Secondly,the fractal value with the transfer time cumulative probability of 85% is determined as the maximum transfer time value.The maximum transfer time of walking,cycling and bus was 11.6min,11.8min and25.6min respectively.Thirdly,considering the impact of different land use on the travel speed,the catchment areas of walking,cycling and bus area are defined based on the cost distance method.Ultimately,the multistage catchment areas including core influence area,secondary influence area and potential influence area are divided by intersecting analysis.Traditional econometric models have the contradiction that many independent variables and multicollinearity problems cannot be considered as both.Numerous factors have influence on railrailway transit demand and cound be divided into four types,i.e.,built environment,external connectivity,multi-mode traffic connection and station attribute.31 candidate influencing factors are collected based on 72 railway stations in Wuhan,China and 19 influencing factors are significantly correlated with railway ridership by bivariate correlation test.A principal component regression(PCR)is proposed to overcome the issue of multicollinearity that widely occurs in multivariate regression analyses for DRM modeling,especially the ordinary least squares regression.Prior to the DRM causal analysis,four principal components are obtained and named commuting trips related factors,elastic trips related factors,regional factors and station attributes.The results show that more than 75% of the total variance can be explained using the four principal components.The first component named factors related to commuting trips,explains approximately 28% of the variance of the original variables.Nineteen determinants have close relationships with MRT demand,with transfer station,the number of jobs,the official building area and the dummy CBD the four most influential factors.Variables in station-area built environment related factors exert more significant impact on MRT ridership than others.The distance to city center has negative association with MRT demand.Traditional econometric models often have the contradiction that many independent variables and multicollinearity problems cannot be considered as both.Due to the numerous influencing factors of rail transit passenger flow,this paper divides it into four aspects: built environment,external connectivity,multi-mode traffic connection and site attribute,and collects 31 candidate influencing factors.Through bivariate correlation test,19 influencing factors that are significantly correlated with rail passenger flow are obtained.Using the principal component regression model as the tool to construct the direct demand forecasting model of track station can eliminate the multicollinarity among the multivariate explanatory variables and realize the dimensionality reduction of many explanatory variables.The results show that many influencing factors can be classified into four principal components,that is,factors related to rigid travel demand,factors related to elastic travel demand,location factor and site attribute.These four principal components account for 80% of the information of the original influencing factors.Among them,the factors related to rigid travel demand account for 56% of the total variance,which occupies a dominant position.From the principal component regression results,VIF of the four principal components are all 1.000,indicating that the regression model basically does not exist multicollinearity.The number of posts,administrative office area,whether the station is a subway transfer station,and whether it is located in the CBD are the four factors that have the largest influence on the passenger flow of the railway station.It is found that the traditional econometric model has spatial autocorrelation of model residuals possibly,which leads to the inaccuracy of the estimation model.This indicates that the railway station ridership may be nonstationary in space.Therefore,the spatial autocorrelation detection of the station ridership and its influencing factors shows that the station ridership has obvious spatial agglomeration.Explanatory variables are determined by the tests of the correlation with passenger volume,multicollinearity between explanatory variables and spatial autocorrelation of explanatory variables.The spatial regression model is selected by Lagrange multiplier test to construct the spatial econometric theoretical model between railway demand and the built environment and site attributes.The results show that there is a significant spatial dependence of railway ridership at the station level.The metro line 2 presents a high-value aggregation,while light rail line 1 and metro line 4 have a low-value aggregation.From a global perspective,the spatial error model has a better fit than the spatial lag model.The number of jobs,official building area,the mixed land use,the number of bus lines and the distance to the CBD have significant influence on the passenger flow of rail stations.The coiffcients of explanatory variables in different catchment areas have heterogeneity.The jobs in the core influence area has a more significant influence than the resident population,and the number of buslines has an obvious promoting effect on the rail passenger flow while an inhibitory effect in the secondary influence area and the potential influence area.The coefficient of the number of schools is negative in the core influence area and positive in the secondary influence area and potential influence area.The influence of mixed land use decreases gradually with the displacement of the core catchment area.The influence of site location factors increases with the distance between the site and the city center.Both traditional econometric model and spatial regression model regard the entire research area as homogeneous,and the relationship between rail passenger flow and surrounding built environment and site attributes is analyzed with a global perspective.These methods ignore the influence of spatial heterogeneity on the prediction of railway demand.In this paper,the forecasting model of railway ridership is built based on the geographic weighted regression.The results show that compared with the least squares regression model and the spatial error model,the geographic weighted regression model has a higher fit.The number of jobs,the official building area,the road length,the number of bus lines and the distance to the city center has the biggest variation in spatial structure.From the perspective of the spatial distribution of influencing factors,the estimated coefficient of influencing factors in Wuchang region is greater than that in Hankou region and Hanyang region,and the coefficient of influencing factors in Hanyang region is generally smaller.The mixed land use along the Yangzi River can promote the passenger flow obviously.Wuguo district is the extreme value area of coefficients of the office building area.The area from the station of Jiedaokou to the optical valley square is the extreme value area of the coefficients of four factors,including the number of posts,the total length of the road,the distance from the rail station to the city center,and the number of bus lines.The DRMs proposed in this paper is a feasible and reliable method for MRT demand forecasting.The findings provide insights for evaluating new metro lines and forecasting the consequences of land use development under the current MRT system.