摘要
为了揭示居民单日活动链与出行时空要素对其出行方式选择的协同影响机制,基于出行调查数据,采用基于密度的噪声应用空间聚类(DBSCAN)方法对个体单日活动链进行时间切片重构与模式识别,提取出包含典型通勤模式(Home-Work-Home,HWH)在内的7类活动链特征.在此基础上,构建轻量级梯度提升机(LightGBM)机器学习模型预测居民出行方式选择,并运用SHAP方法解析关键影响因素的作用机理.研究发现:1)出行距离和时长是影响方式选择的核心决定因素;2)HWH活动链与时空要素存在显著交互效应,表现为通勤群体在短时长(<40 min)中对公共交通的偏好程度显著低于其他群体,而在20~40 min出行时长区间内对电动自行车表现出更强的选择偏好,证实了活动链类型驱动的出行行为异质性特征.
The synergistic impact mechanism of residents’daily activity patterns and spatio-temporal travel factors on their travel mode choice was investigated.Based on travel survey data,the density-based spatial clustering of applications with noise(DBSCAN)method was employed to perform time-slice reconstruction and pattern recognition of individuals’daily activity patterns,extracting 7 activity pattern characteristics,including the typical Home-Work-Home(HWH)commute pattern.Subsequently,a light gradient boosting machine(LightGBM)machine learning model was constructed to predict residents’travel mode choice,with the SHapley Additive exPlanations(SHAP)method used to interpret the mechanisms of key influencing factors.The findings revealed the following two key results:1)Travel distance and duration were the core determinants of mode choice.2)There was a significant interaction between the HWH pattern and spatio-temporal factors:compared to other groups,HWH commuters showed a lower preference for public transportation for trips under 40 min,yet a stronger preference for electric bicycles for trips between 20 and 40 min.These confirmed the activity pattern type-driven heterogeneity of travel behavior.
作者
沈向诚
孙轶琳
谌淑杰
SHEN Xiangcheng;SUN Yilin;CHEN Shujie(College of Civil Engineering and Architecture,Zhejiang University,Hangzhou 310058,China;Polytechnic Institute of Zhejiang University,Hangzhou 310058,China)
出处
《浙江大学学报(工学版)》
北大核心
2026年第4期865-875,共11页
Journal of Zhejiang University(Engineering Science)
基金
国家自然科学基金重点项目(52131202)
浙江省“尖兵”“领雁”研发攻关计划资助项目(2024C01214).
关键词
活动链
出行方式选择
机器学习
通勤活动
DBSCAN聚类
activity pattern
travel mode choice
machine learning
commuting activities
DBSCAN clustering
