The invention discloses a collaborative traffic signal control method for enhancing multi-agent characterization based on comparative learning. The method comprises the following steps: acquiring traffic information of multiple intersections of a road network; setting parameters based on a Markov decision process; each intersection is used as a reinforcement learning agent, and an agent strategy network with characterization capability is trained based on shared neighbor parameters; and updating the intelligent agent strategy network based on comparative learning, and outputting a final control strategy. According to the method, the characterization capability of the reinforcement learning agent is enhanced through comparative learning, so that the characteristic representation which better conforms to the characteristics of the local intersection is sensitively captured, and a traffic signal control scheme with better cooperative consciousness is provided.
本发明公开了一种基于对比学习增强多智能体表征的协同交通信号控制方法,包括获取道路网络的多交叉口的交通信息;基于马尔科夫决策过程进行参数设定;每个交叉口作为一个强化学习智能体,基于共享邻居参数训练具有表征能力的智能体策略网络;基于对比学习更新智能体策略网络,并输出最终的控制策略。本发明通过对比学习增强强化学习智能体表征能力,以敏锐的捕捉更符合本地交叉口特点的特征表示,提供更具合作意识的交通信号控制方案。
Cooperative traffic signal control method for enhancing multi-agent characterization based on comparative learning
基于对比学习增强多智能体表征的协同交通信号控制方法
2024-06-14
Patent
Electronic Resource
Chinese
Distributed multi-agent traffic signal cooperative control method
European Patent Office | 2025
|European Patent Office | 2024
|European Patent Office | 2024
|