The invention discloses a traffic signal control method based on spatial-temporal feature extraction and reinforcement learning, which is used for multi-intersection traffic signal cooperative control. The method comprises the following steps: learning an interdependence influence degree relationship between intersections, and dynamically dividing an intersection region based on a space-time influence degree; a space-time diagram modeling module, a feature embedding module, a time sequence Transform feature extraction module and a space diagram Transform feature aggregation module are constructed, all intersections in each time slot are modeled into space-time diagrams, and space-time features of the intersections are extracted; the intersection is used as an intelligent agent, an action strategy network and an action duration time strategy network are set, and traffic signal cooperative control is carried out based on repeated action multi-agent deep reinforcement learning. According to the method, effective intersection features can be extracted, the action and the time granularity are adaptively decided according to the fluctuating real-time traffic state, regional cooperative control of intersection signals is achieved, and the real-time performance and the safety of signal decision are improved.
本发明公开了一种基于时空特征提取和强化学习的交通信号控制方法,用于多路口交通信号协同控制。本发明方法包括:学习路口之间的相互依赖影响程度关系,基于时空影响程度对路口区域进行动态划分;构建时空图建模模块、特征嵌入模块、时序Transformer特征提取模块和空间图Transformer特征聚合模块,将每一时隙所有路口建模为时空图,提取路口时空特征;将路口作为智能体,设置动作策略网络和动作持续时间策略网络,基于重复动作多智能体深度强化学习进行交通信号协同控制。本发明方法能提取有效的路口特征,根据波动的实时交通状态自适应决策动作和时间粒度,实现路口信号的区域协同控制,提升了信号决策的实时性和安全性。
Traffic signal control method based on spatial-temporal feature extraction and reinforcement learning
一种基于时空特征提取和强化学习的交通信号控制方法
28.05.2024
Patent
Elektronische Ressource
Chinesisch
Spatial and temporal feature extraction method based on traffic data
Europäisches Patentamt | 2024
|Reinforcement learning-based traffic signal control
IEEE | 2024
|Traffic signal control method based on reinforcement learning
Europäisches Patentamt | 2023
|Deep Reinforcement Learning-based Traffic Signal Control
IEEE | 2023
|