城市交通信号局部博弈交互下的学习协调控制

智能交通信号控制系统是智慧交通系统（intelligent traffic system，ITS）的重要组成部分，为形成安全高效的交通环境提供实时服务. 然而，现有自适应交通信号控制方法因通信受限难以满足复杂多变的交通需求. 针对通信时延长和信号灯有效利用率低的难题，提出一种基于边缘计算的异步决策的多智能体交通信号自适应协调方法（adaptive coordination method，ADM）. 该方法基于提出的端—边—云架构实现实时采集环境信息，将异步通信引入强化学习的多智能体协调过程，设计一种多智能体之间使用不同决策周期的异步决策机制. 实验结果表明边缘计算技术为高实时性要求的交通信号控制场景提供一种良好的解决思路，此外，相较于固定配时和独立决策的Q学习决策方法IQA（independent Q-learning decision algorithm）而言，ADM方法基于异步决策机制和邻居信息库实现智能体之间的协作，达到降低车辆平均等待长度及提高交叉口时间利用率的目标.

A unified framework for vehicle rerouting and traffic light control to reduce traffic congestion

The intelligent traffic signal control system is a component of the intelligent traffic system (ITS), offering real-time services for the creation of a safe and efficient traffic environment. However, due to restricted communication, conventional adaptive traffic signal-controlled methods are unable to fulfill the complex and changing traffic requirements. A multi-agent adaptive coordination method (ADM) based on asynchronous decision-making and edge computing is presented to address the issues of communication delay and a decrease in signal utilization. Firstly, the end-side-cloud architecture is proposed for real-time environmental information collection and related processing. Then, to enhance the agent coordination process, asynchronous communication is implemented. An approach for calculating the decision cycle of the agent is presented, and an asynchronous decision mechanism employing multiple agents’ decision cycles is devised. The experimental results show that edge computing technology provides a good solution for traffic signal control scenarios with high real-time requirements. In addition, compared with the fixed time (FT) and independent Q-learning decision algorithm (IQA), ADM achieves collaboration among the agents based on the asynchronous decision mechanism and the neighbor information base, and reduces the average vehicle waiting length and improves intersection time utilization.