Abstract: The arrival unpunctuality of demand- responsive bus transit seriously reduced the service level and the passengers' willingness to choose public transit. This paper studies the dynamic bus routing problem of demandresponsive feeder transit (DRFT) considering stochastic bus arrival time. A mathematical model is developed by optimizing bus routes to find the cost-optimal transit service, in which the total cost, consisting of operator cost, passenger travel time cost, and passenger waiting time cost is minimized. The innovation of the model lies in allowing passengers to submit real-time travel demand during the operation and defining the bus arrival time following a known distribution to describe its stochasticity. A heuristic combining genetic algorithm and neighborhood search was proposed. The proposed algorithm hybridizes the global search of a genetic algorithm with the local search of a neighborhood search algorithm. The validity and advance of the proposed algorithm are verified by the experimental test analysis. Finally, the results based on the experiment on Yanpingmen subway station in Xi'an City show that the consideration of stochastic bus arrival time could reduce the passenger waiting time and the total cost.

Key words: urban traffic, demand- responsive feeder transit, genetic algorithm, neighborhood search, real-time requests, stochastic bus arrival time

摘要： 车辆到站时间的不准时性严重影响着需求响应型公交的服务水平和乘客选择公共交通的出行意愿，因此，本文对考虑车辆随机到站时间的动态需求响应型接驳公交线路优化问题进行研究。以运营商成本、乘客乘车时间成本、乘客等待时间成本组成的系统总成本最小为目标建立数学模型，通过优化车辆路径寻求系统总成本最优的需求响应型接驳公交服务方案，其创新之处在于，在服务过程中允许乘客提交实时出行需求；定义车辆到站时间服从已知分布以描述其随机性。提出一种遗传算法和邻域搜索相结合的启发式算法对模型进行求解，该算法融合了遗传算法的全局搜索优势和邻域搜索的局部搜索能力，通过算例测试分析对本文算法的有效性及先进性进行验证。最后，基于西安市延平门地铁站设计数值实验，结果表明，考虑车辆随机到站时间可以在一定程度上减少乘客时间成本和系统总成本。

关键词: 城市交通, 需求响应型接驳公交, 遗传算法, 邻域搜索, 实时需求, 车辆随机到站时间