| 摘要: |
| 随着城市通勤压力上升，电动自行
车因其灵活性强、适用范围广，逐渐成为居
民日常通勤及接驳换乘工具。本文以济南市
中心城区为研究区域，基于2019 年居民出行
调查数据，运用梯度提升决策树（GBDT）模
型，从居住地与就业地双重视角出发，系统
分析建成环境因素对电动自行车选择概率的
非线性关系及阈值效应。研究发现：第一，
通勤距离是影响选择概率的关键变量，呈倒U
型关系，在2 300 m处达到峰值（55%）；第
二，适度的功能复合、高连通性、适度绿地
配比的空间布局更有利于促进电动自行车通
勤出行；第三，收入水平对出行行为影响显
著，低收入群体选择概率最高（63.9%），高
收入群体最低（19.5%）；第四，不同收入群
体对环境要素敏感性存在差异：低收入群体
主要受到居住地建成环境的影响，中等收入
群体则受到职住两地的双重影响，而高收入
群体更敏感于就业地的建成环境。研究结果
能为优化中短距离通勤空间结构，推动城市
交通系统绿色、高效与包容转型提供参考。 |
| 关键词: 电动自行车 建成环境 通勤选择
概率 社会分异 梯度提升决策树 |
| DOI：10.13791/j.cnki.hsfwest.20250429001 |
| 分类号: |
| 基金项目:国家自然科学基金青年科学基金项目（52408077）；教育部人文社会科学研究规划项目（24YJCHZ234、24YJA890051）；
山东省自然科学基金青年科学基金项目（ZR2023QE242）；山东省社会科学规划研究项目（23DSHJ07、24DSHJ06） |
|
| Non-liner influence of the built environment at residential and workplace locations one-bike commuting probability ： A perspective of social differentiation |
|
QIU Ning,YANG Chuanzheng,HAN Xinyu,JIANG Yuxiao,ZHANG Zhiwei
|
| Abstract: |
| With the intensification of urban commuting pressures, electric bicycles (e-bikes) have
emerged as a vital supplement to conventional transportation modes due to their flexibility, costefficiency,
and low environmental impact. Compared to traditional bicycles, e-bikes offer improved
efficiency and convenience, making them particularly suitable for short- and medium-distance travel.
While recent national and local policies have encouraged their adoption, there remains a notable
research gap concerning the systematic spatial planning and environmental adaptation for e-bike travel.
This study investigates the factors influencing e-bike commuting behavior in the main urban area of
Jinan, China. Drawing on 2019 household travel survey data encompassing 15 990 households and
over 38 000 individuals, the study integrates multi-source geospatial data—including land use, POIs,
road networks, and real estate prices—and constructs built environment and socioeconomic indicators
for both residential and workplace locations. Using a Gradient Boosting Decision Tree (GBDT) model,
the study quantifies the nonlinear impacts and threshold effects of built environment features on the
probability of choosing e-bike commuting and further examines how these effects vary among income
groups. Key findings are summarized as follows: Distance-Dependent Mode Choice. E-bike
commuting exhibits a clear nonlinear response to commuting distance. The probability of selecting an
e-bike peaks at around 2 300 meters (with a maximum probability of 55%), indicating strong
suitability for mid-range commuting (1 000–3 500 meters). However, this probability declines sharply
beyond 3 500 meters and is nearly negligible beyond 20 kilometers. In contrast, cars dominate for
long-distance commuting (>6 000 meters), while walking is most prevalent under 1 000 meters.Built
Environment Effects. The impact of built environment characteristics on commuting decisions is
complex and nonlinear. Residential location factors—particularly building density, land-use mix,
public service land ratio, and road network density—exert more substantial influence compared to
workplace factors. High residential building density, for instance, enhances e-bike usability by
ensuring population clustering and proximity to service facilities. Conversely, extremely dense
commercial or road networks, particularly in workplace zones, may deter e-bike usage due to
congestion, safety risks, and limited parking availability. Socioeconomic Differentiation. There is
significant heterogeneity in commuting behavior responses across income groups. Low-income
individuals are more sensitive to spatial accessibility and heavily reliant on e-bikes due to limited
access to private vehicles or public transport. Their commuting choices are predominantly shaped by
residential environment features such as the availability of public services and road connectivity.Middle-income groups are influenced by both residential and workplace built environments, reflecting more dispersed employment locations and greater interdistrict
mobility. High-income individuals prioritize commuting efficiency and comfort, showing greater responsiveness to road infrastructure quality and
multimodal transport integration (e. g., proximity to public transport nodes). Their residential preferences are aligned with conventional residential areas
characterized by clear functional zoning. Built Environment Threshold Effects. GBDT model outputs reveal specific tipping points for built environment
variables. For example, residential green space ratios exceeding 15% significantly boost e-bike commuting probability, while public service land ratios above
30% show a U-shaped relationship. Excessive land-use mixing (e.g., mix indices >1.0) may reduce commuting necessity, thus diminishing e-bike attractiveness.
Policy Implications. The findings underscore the need for differentiated spatial interventions and transportation policies. To encourage e-bike commuting, urban
planners should focus on enhancing building density, service facility accessibility, and road connectivity in residential neighborhoods. In employment zones,
balanced land-use planning—avoiding overly centralized commercial or industrial zoning—is essential to maintain e-bike feasibility. Furthermore, policy
measures must address commuting equity by prioritizing infrastructure development in low-income areas and tailoring multimodal transit integration strategies
for high-income groups.In conclusion, e-bike commuting is well-suited for short- to mid-distance urban travel and offers a promising path toward sustainable
mobility. However, its adoption is constrained by spatial, infrastructural, and socioeconomic factors that exhibit complex, nonlinear dynamics. This study
contributes to the evolving discourse on green urban mobility by integrating machine learning with spatial planning theory, offering a robust empirical basis for
refined, equity-oriented transportation planning.Nevertheless, some limitations persist. The study focuses on a single temporal snapshot and does not account
for temporal dynamics in commuting behavior. Additionally, potential confounding variables—such as lifestyle preferences and employer-provided transport
services—were not fully controlled. Future research could leverage multi-temporal datasets to model commuting behaviors dynamically and extend the analysis
to other cities for broader applicability. |
| Key words: E-bike built environment commuting choice probability social differentiation Gradient Boosting Decision Tree (GBDT) |
