基于贝叶斯网络的城市公交动态调度决策方法

项目名称： 基于贝叶斯网络的城市公交动态调度决策方法

项目编号： No.61503201

项目类型： 青年科学基金项目

立项/批准年度： 2016

项目学科： 自动化技术、计算机技术

项目作者： 魏明

作者单位： 南通大学

项目金额： 19万元

中文摘要： 城市公交处于动态不确定环境中，现实中的众多不确定因素会引起客流或行驶时间变化，这可能导致当前时刻表、车辆和驾驶员排班方案失效，现有研究不能很好地解决随机干扰致使这些调度方案需要动态实时调整问题。.针对现有研究的不足，本课题利用贝叶斯网络探索公交动态调度问题。将该问题划分为公交环境感知、随机干扰对调度方案的影响评估、调度方案动态调整等。首先，利用贝叶斯网络分别刻画上述决策过程中变量之间的因果关系；其次，集成它们于一体，描述随机干扰引起时刻表、车辆和驾驶员排班在实际系统中的状态转移过程。当交通事件发生时，评估它们致使当前调度方案失效的原因及其调整方向，借鉴调度人员的先验知识，结合历史和实时公交运营数据，在线学习模型结构和参数，利用模型推理算法动态生成下一时刻的可行调度方案，达到自主决策的目的。.将该成果开发为智能公交调度决策系统，以南通某公交公司为试点，进行测试、完善和应用。

中文关键词： 常规公交；公交调度；动态调度；实时调度；贝叶斯网络

英文摘要： Bus scheduling, in a dynamic uncertain environment, is influenced by uncertainties to change their passenger flow or travel time. This may lead to failure in executing current timetable, vehicle and driver scheduling plan. Existing research cannot easily deal with such a real-time bus schedule adjustment problem..Given the lack of existing research, this study uses Bayesian network to explore bus dynamic scheduling problem. It is divided into perception of transit environment, assessment of traffic incident‘s influence on bus scheme, and its dynamic adjustment. Firstly, Bayesian network is used to portray causal relationship between variables in above decision-making process. Secondly, they are integrated as a whole to describe the state transition caused by the random interference between timetable, vehicle and driver scheme in a real system. When a traffic incident occurs, assess reasons and directions of adjustment for it causing current scheme’s failure. Dispatchers’ prior knowledge, combined with historical and real-time data, is used to learn model structure and parameters online. On this basis, a feasible solution is dynamically generated using the model inference algorithms. Accordingly, our work achieves the purpose of independent decision-making..An intelligent dynamic bus scheduling decision system is explored to test our application in Nantong Public Transport Company.

英文关键词： Conventional Bus;Bus Schedulling;Dynamic Scheduling;Real-time Scheduling;Bayesian Networks