【作者】 罗鹏；

【导师】 杨鸿泰；

【作者基本信息】 西南交通大学 ， 资源与环境（专业学位）， 2024， 硕士

【摘要】 近年来,随着互联网和通信技术的快速发展,诞生了许多新型的出行方式,例如自动驾驶汽车、网约车出行等。网约车作为一种新型的共享出行方式,因其服务方式灵活、服务质量相对较高等特点而被大众广泛接受。然而,越来越多的网约车出行也带来了不利影响,例如加剧了空气污染和交通拥堵。因此,拼车服务顺势而生,出行者可以以较低的出行费用与他人共享同一行程(车辆),为出行者提供了一种经济的出行选择;另一方面,通过减少出行的车辆里程,缓解了空气污染和交通拥堵。基于以上原因,拼车受到了政府的鼓励和推广。为回答哪些因素会影响出行者的拼车行为,部分研究已经探讨了建成环境与拼车行为的关系,但在控制出行起点和出行终点特征的情况下,很少有研究探索费用折扣这一重要因素对拼车行为(是否拼车)的影响。因此,本文通过分析芝加哥2019年前五个月的网约车出行数据来探讨这一话题,全面分析这些因素对拼车行为的影响机制。本文的主要内容包括:(1)拼车行为的时空特征分析(2)以出行起终点所在的人口普查区之间的拼车率(OD对拼车率)作为因变量,引入半参数广义可加模型,从宏观层面(集计模型)刻画影响因素与拼车率之间的非线性关系(3)以每一次出行记录中是否拼车(0-1变量)作为因变量(非集计模型),使用梯度提升决策树模型,从微观个人层面捕捉影响因素与拼车意愿之间的非线性关系(4)费用的敏感性分析及具有拼车提高潜力区域的识别。半参数广义可加模型固定效应部分结果表明,年轻人比例、无车家庭比例等与拼车率表现出正相关关系。受高等教育比例、收入中位数等与拼车率呈现负相关关系;平滑函数部分表明人口密度、岗位密度、公共交通密度、低收入比例、费用折扣以及出行距离等变量与拼车率存在着显著的非线性关系。利用梯度提升决策树的部份依赖图捕捉了费用折扣、建成环境与拼车行为的非线性关系。研究结果发现,建成环境类因素均观察到明显的阈值效应。例如人口密度在达到22000人/平方英里之后,对个人拼车意愿的影响逐渐平稳;而岗位密度达到50000个/平方英里之后,对拼车意愿的负向影响突然加剧。值得注意的是费用折扣的阈值效应出现在0.23-0.4这个区间,拼车意愿表现出明显的提升。以上结果可以帮助交通规划者和政府机构提升拼车服务,并为交通网络公司制定适当的费用折扣提供指导。更多还原

【Abstract】 In recent years,with the rapid development of the Internet and communication technologies,various new modes of transportation have emerged,such as automatic driving and ride-hailing.As a new type of shared travel mode,ride-hailing has been widely accepted by the public because of its flexible service mode and service quality.However,the increasing number of ride-hailing trips has also brought adverse effects,such as increasing air pollution and traffic congestion.As a result,ridesplitting services have emerged in response to this trend.Travelers can share the same trip(vehicle)with others at a lower travel cost,which provides an economical transportation option for travelers;on the other hand,it alleviates air pollution and traffic congestion by reducing the number of vehicle miles traveled,and for the above reasons,ridesplitting has been encouraged and promoted by government agencies.To answer the question of which factors affect travelers’ridesplitting behavior,some studies have explored the relationship between the built environment and ridesplitting behavior,but few studies have explored the effect of the important factor of fare discounts on ridesplitting behavior(whether to ridesplitting or not),controlling for trip origin and trip destination characteristics.Therefore,this paper explores this topic by analyzing ride-hailing trip data from January to May 2019 in Chicago to analyze the factors that influence ridesplitting intentions.The main contents include:(1)a spatio-temporal characterization of ridesplitting behavior(2)using ridesplitting adoption rate between trip origin and trip destination(OD pairs ridesplitting adoption rate)as the dependent variable,a semi-parametric generalized additive model(aggregate model)was used to explore the factors affecting ridesplitting adoption rate at the macro level(3)using whether or not to ridesplitting(0-1 variable)in each trip record as the dependent variable(disaggregate model),Gradient Boosting Decision Tree model is used to explore the key factors influencing whether or not a traveler chooses to ridesplitting at the micro-individual level,and to capture the non-linear relationship between them(4)fare sensitivity analysis and identification of areas with potential for ridesplitting enhancement.The results of the fixed effects section of the semi-parametric generalized additive model show that the proportion of young people,the proportion of carless households,etc.positively affect ridesplitting adoption rates;the proportion of higher education,the median income,etc.show a negative correlation with ridesplitting adoption rates;the smoothing function part shows that the variables of population density,job density,public transportation density,low-income proportion,cost discount,and travel distance have a significant nonlinear relationship with ridesplitting adoption rates.GBDT was used to capture the nonlinear relationship between trip characteristic variables,built environment variables and ridesplitting behavior.The results of the study found that significant threshold effects were observed for all factors in the built environment category.For example,the effect of population density on an individual’s willingness to ridesplitting gradually smoothes out after reaching 22,000 people/square mile,while the negative effect on willingness to ridesplitting suddenly intensifies after job density reaches 50,000 people/square mile.It is worth noting that the threshold effect of fare discount appears in the interval of 0.23-0.4,where ridesplitting willingness shows a significant increase.The above results can assist transportation planners and government agencies in identifying areas where ridesplitting services need to be improved and provide guidance to transportation network companies in developing appropriate fare discounts.更多还原