Journal of Transportation Systems Engineering and Information Technology ›› 2022, Vol. 22 ›› Issue (3): 45-52.DOI: 10.16097/j.cnki.1009-6744.2022.03.006

Previous Articles     Next Articles

Dynamic Fleet Management of Shared Autonomous Vehicles with Rolling Horizon Optimization

CHEN Yao1 , BAI Yun* 1 , ZHANG An-ying2 , MAO Bao-hua1 , CHEN Shao-kuan1   

  1. 1. Key Laboratory of Transport Industry of Big Data Application Technologies for Comprehensive Transport, Beijing Jiaotong University, Beijing 100044, China; 2. Guangdong Provincial Transport Planning & Research Center, Guangzhou 510101, China
  • Received:2022-03-09 Revised:2022-03-24 Accepted:2022-03-31 Online:2022-06-25 Published:2022-06-22
  • Supported by:
    National Natural Science Foundation of China (72101019, 71971021);Fundamental Research Funds for the Central Universities (2021RC228)。


陈垚1,柏赟* 1,张安英2,毛保华1,陈绍宽1   

  1. 1. 北京交通大学,综合交通运输大数据行业重点实验室,北京 100044;2. 广东省交通运输规划研究中心,广州 510101
  • 作者简介:陈垚(1993- ),男,江西抚州人,讲师,博士。
  • 基金资助:

Abstract: The shared autonomous vehicle (SAV) is an essential component in future urban transportation systems. This paper investigates an optimization approach to the dynamic operationof a SAV fleet with stochastic demand. A timespace network is first constructed to characterize the fleet management problem. Different types of time-space arcs are generated to indicate the vehicle-trip assignment and empty vehicle relocation. Under the framework of approximated dynamic programming, this paper develops a mathematic programming model to maximize the operational profit, in which the flow of nodes is taken as vehicle states and the flow of arcs is taken as decision variables. The rolling horizon optimization, also referred as lookahead policy, is designed for the optimization problem. A stochastic program with a lookahead horizon is developed and solved by the CPLEX solver. A numerical case study is performed with the Sioux Falls network. The rolling horizon optimization approach can provide effective operational decisions of dynamic fleet management. Considering the computational time limit, a long lookahead horizon with a medium- size sample would produce better optimization results. The objective of maximizing the operational benefit while minimizing the passenger waiting time would also result in more effective decisions of the dynamic fleet management.

Key words: urban traffic, shared mobility, autonomous vehicles, rolling horizon, dynamic program, stochastic demand

摘要: 共享自动驾驶汽车被视为未来城市交通系统的重要组成部分。本文考虑随机订单需求研究共享自动驾驶汽车的动态调度优化方法。通过建立车辆调度时空网络,分别针对订单分配与空车移位生成车辆运行时间弧,提出车辆调度问题的刻画方法。基于马尔科夫决策框架,以时空节点流量为状态,以时空弧流量为决策变量,建立最大化系统净收益的车辆动态调度优化模型。 采取滚动时域优化思想,建立含前视时间窗的随机规划模型,并利用CPLEX优化引擎,滚动求解车辆动态调度决策结果。Sioux Falls网络算例结果表明,滚动时域优化方法可保证车辆动态调度决策效果,提升系统运营效率。在计算时间限制下,滚动时域方法应优先采用长时间窗中等规模 样本。在最大化系统净收益的同时进一步最小化乘客等待时间,可有效提升车辆动态调度决策效果。

关键词: 城市交通, 共享出行, 自动驾驶, 滚动时域, 动态规划, 随机需求

CLC Number: