建成环境对轨道交通客流非线性影响的空间特征

Spatial Patterns of Nonlinear Effects of Built Environment on Beijing Subway Ridership

机器学习模型是研究建成环境对城市轨道交通车站客流影响的非线性方法,但既有的解释工具对该类模型结果空间特征的分析仍不够充分。基于多源位置大数据提出城市轨道交通建成环境的多维度指标及其计算方法,采用极限梯度提升决策树和局部回归分析技术,构建样本的κ指标以分析建成环境对客流非线性影响的空间特征。针对北京地铁的案例研究表明:建成环境对城市轨道交通车站出站客流的非线性影响差异显著,就业密度、公共交通可达性和用地混合度等排序前三的解释变量的重要度高达42.51%。这3个解释变量的κ指标空间分布呈现显著的空间异质性,表明客流对建成环境的依赖关系具有空间非平稳性。上述建成环境对客流非线性影响的空间效应说明:城市轨道交通车站周边用地开发不仅宜采取因地制宜的差异化发展策略,而且还需合理确定不同区域车站资源开发配置的下限值,激活建成环境因素的阈值效应以改善客流效果。

Machine learning is a nonlinear approach to examine the effects of built environment on station-level ridership of urban rail transit.But existing interpretation approach are not able to analyze the spatial patterns of results derived from machine learning models.This paper first utilizes multi-source location-based big data to quantify the indicators of built environment of urban rail transit,and then uses theκindictor to analyze the spatial distribution of nonlinear effect of built environment on urban rail transit passenger flow based on the extreme gradient boosting incorporating local regression technique.The case study of Beijing subway shows that the nonlinear effects of built environment factors on egress flow of urban rail transit is significantly different.The top three indictors of built environment,including employment density,public accessibility,and mixed land use,account for 42.51%of total importance of variables.The spatial distributions ofκindictor of these three built environment factors show significant spatial heterogeneity,indicating the spatial non-stationary relationship between the passenger flow and built environment.The nonlinear spatial patterns indicate that the development of urban rail transit stations should not only adopt differentiated policies and strategies in different areas,but should also determine reasonable lower limit of resource to activate threshold effect of built environment,which would help to increase the ridership of urban rail transit.