Abstract: Revealing the mechanism of the influence of built environment and weather on passengers’ bus commuting time is the key to optimize passengers’ commuting time and improve bus sharing rate. Taking Weinan city in Shaanxi as an example, the variables such as passengers’ commuting time during evening peak hours, weather, and the built environment of the origin and destination were extracted through the fusion and mining of multiple sources of spatiotemporal big data. The CatBoost model and SHAP were used to analyze the feature importance and nonlinearity influence of the built environment and the weather on passengers’ bus commuting time during the evening peak hours. The research results show that CatBoost model has good fitting and prediction performance, which is significantly better than XGBoost, KNN, LR and Bayesian Ridge. The significance influencing factors in descending order are: origin science and education density, origin residential density, destination life service density, origin diversity, origin road density, destination residential density, destination diversity, temperature, and precipitation. Among them, the residential density of origin and destination, origin diversity, origin road density have a significant negative impact on the commuting time of passengers during the evening rush hour, while the increase of destination diversity and the rise of the temperature will potentially increase the bus commuting time of passengers during the evening rush hour. The origin science and education density, destination life service density have a significant nonlinear influence on the commuting time of residents.

Key words: traffic and transportation engineering; urban traffic; nonlinearity; CatBoost; SHAP; bus commuting time; built environment

摘要： 揭示建成环境和天气对乘客公交通勤时间的影响机理是本源性优化乘客通勤时间、提升公交分担率的关键。以陕西省渭南市为例，通过多源时空大数据的融合与挖掘，提取乘客晚高峰公交通勤时间、天气和起终点建成环境变量，应用CatBoost（category boosting）模型和SHAP（shapley additive explanation）解析天气和起终点建成环境对乘客晚高峰公交通勤时间的特征重要度和非线性影响。研究结果表明：CatBoost模型具有很好的拟合和预测性能，明显优于XGBoost、KNN、LR和Bayesian Ridge模型；显著性影响因素依次为，起点科教密度、起点居住密度、终点生活服务密度、起点多样性、起点道路密度、终点居住密度、终点多样性、温度和降水；其中，起终点居住密度、起点多样性、起点道路密度对乘客晚高峰公交通勤时间有显著的负向影响，而终点多样性的增加以及温度的上升会潜在增加乘客晚高峰公交通勤时间；起点科教密度、终点生活服务密度对通勤时间有显著的非线性影响。

关键词: 交通运输工程；城市交通；非线性；CatBoost；SHAP；公交通勤时间；建成环境