空间异质下地铁建成环境与站点覆盖客流吸引度关系研究  

Relationship Between Built Environment of Metro Station and Passenger Attraction Considering Spatial Heterogeneity

作　　者：陈红[1] 李晨光[1] 王铎 段超杰 姚振兴 CHEN Hong;LI Chenguang;WANG Duo;DUAN Chaojie;YAO Zhenxing(School of Transportation Engineering,Chang'an University,Xi'an 710064,China)

机构地区：[1]长安大学,运输工程学院,西安710064

出　　处：《交通运输系统工程与信息》2024年第4期253-262,共10页Journal of Transportation Systems Engineering and Information Technology

基　　金：国家自然科学基金青年科学基金(52002030)。

摘　　要：机器学习模型广泛应用于探究建成环境与客流的交互关系。然而,机器学习考虑的是全局关系,无法捕捉空间变化,为解决这一问题,本文从密度、多样性、设计、目的地可达性、可获得性和网络连通性等方面构建了11种建成环境指标,提出一种轻量级梯度提升机(LightGBM)与地理加权回归(GWR)的集成综合分析模型(SLightGBM),以探究建成环境对站点覆盖客流吸引度的空间异质性和非线性影响,并将该模型与LightGBM,普通最小二乘法(OLS)和GWR进行比较,以揭示SLightGBM模型在回归效果上的优势。针对西安市的研究结果表明:SLightGBM模型的R2、MAE与RMSE分别达到了0.68、8379.16和11797.19,显著优于对比模型;建成环境因素存在空间异质性,工作人口密度和公交站点密度在中心区域最为重要,而餐饮密度在南部区域更显著;工作人口密度和餐饮密度与客流吸引度呈正相关,而最小换乘次数与客流吸引度呈负相关关系,且具有明显的协同作用。研究表明,影响因子空间差异性和阈值分析结果对于指导城市建设和改善公共交通系统具有重要的启示意义。Machine learning models have been extensively applied in exploring the interaction between the built environment and passenger flow.However,machine learning primarily considers global relationships and fails to capture spatial variations.To address this issue,this paper defined 11 built environment variables from the aspects of density,diversity,design,destination accessibility,availability,and network connectivity.The study proposed an integrated ensemble analysis model,SLightGBM,combining Light Gradient Boosting Machine(LightGBM)with Geographically Weighted Regression(GWR),to investigate the spatial heterogeneity and nonlinear impact of the built environment on the attractiveness of station coverage.The SLightGBM model was compared with the LightGBM,Ordinary Least Squares(OLS),and GWR to demonstrate its regression superiority.The results from Xi'an city indicate that:(1)The SLightGBM model showed better performance than other models,with 2 R value of 0.68,MAE of 8379.16,and RMSE of 11797.19.(2)The factors of the built environment vary across spaces.The densities of the employment and bus stops are most important in central areas,whereas the density of restaurants is more prominent in the southern regions.(3)Higher employment and restaurants densities are positively correlated with the attractiveness of metro ridership,while the minimum transfer times are negatively correlated with the attractiveness of metro ridership,showing a strong combined effect.This study indicates the importance of understanding spatial differences and threshold effects of these factors in urban planning and public transport system improvement.

关 键 词：城市交通 空间特性 SLightGBM 地铁客流 阈值分析 

分 类 号：U268.6[机械工程—车辆工程]