基于序优化和遗传算法的大规模交通系统协调控制研究

At almost all big cities around the world, the traffic congestion is a notorious problem, which causes a great waste of the social resources. People are very concerned with how to improve the traffic situation without changing the infrastructures. Given the road network and the traffic demands, how much better can we do by improving the management and control methods? Given the infrastructure limitation, how many vehicles should be allowed to go on road? For a large road network, there is little quantitative research for these problems. We propose to use Genetic Algorithms and the Ordinal Optimization method to solve single and multiple objective coordinated control problems for a road network as large as Beijing, based on the multi-agent model running on the computational experiments platform built with Graphics Processing Units (GPU), which are a powerful parallel computing tools. We will compare with popular methods of traffic control and show how much better we can achieve. The results can provide useful references for making decisions and help take better use of resources.

在世界的各大城市，交通拥堵都是严重问题，极大的浪费了社会资源。在不改变基建设施的情况下，如何通过提高管控水平来改善交通状况是一个重要问题。给定交通路网和交通出行的情况下，通过改变管控方法究竟能够得到多好的交通状况？或者说，在给定基础设施的情况下，究竟该限制多少车辆出行？在大规模路网上，针对这一问题尚缺乏定量研究。本项目在基于大规模并行计算工具--图形处理器建立的交通系统多智能体模型计算实验平台上，拟对北京这一规模的路网进行整个路面交通的微观仿真，对单目标和多目标协调控制问题，结合动态控制子区分区方法，用遗传算法和专用于大规模系统优化的序优化方法进行求解，并和现行的控制方法作对比。通过这一研究，希望能够定量分析路网通行能力的上限以及改进空间。该结果能够为限行限号等决策提供重要参考依据，有利于更好地利用资源。