The invention discloses a deep reinforcement learning traffic signal control method based on an attention mechanism, and the method comprises the steps: constructing a traffic road network model, and obtaining the vehicle traffic information of each lane of each road; analyzing a current road network model, establishing a multi-agent deep learning framework taking each intersection as an agent, and setting an abstract definition and a set; constructing a Q learning off-line strategy and a gamma attention reward strategy based on a D3QN reinforcement learning basic network structure by adopting a decentration thought; based on a gamma attention reward strategy, designing a revised playback data buffer layer algorithm based on an attention mechanism; simulation data and actual detection traffic data are selected, the number of neighbors of each intersection agent is initialized by adopting a hyper-parameter neighbour action range in a Colight algorithm, simulation iteration is carried out according to the actual traffic flow data, an optimal decision simulation result is quickly obtained, and the problem of urban traffic congestion of multiple intersections is solved.
本发明公开了一种基于注意力机制的深度强化学习交通信号控制方法,构建交通道路路网模型,获得各个道路每个车道的车辆交通信息;分析当前道路路网模型,建立以各个交叉口为代理的多智能体深度学习框架,设定抽象定义及集合;采用去中心化思想,基于D3QN增强学习基础网络结构,构建Q学习离线策略以及γ注意力奖励策略;基于γ注意力奖励策略,设计基于注意力机制的修正回放数据缓冲层算法;选取模拟数据和实际检测交通数据,采用Colight算法中超参数邻居作用域确定各个交叉口代理的邻居数目初始化,根据交通流实际数据带入进行仿真迭代,快速得到最优决策仿真结果,解决了多交叉口的城市交通拥堵的问题。
Deep reinforcement learning traffic signal control method based on attention mechanism
一种基于注意力机制的深度强化学习交通信号控制方法
2024-03-22
Patent
Electronic Resource
Chinese
Deep reinforcement learning traffic signal decision system and method based on attention mechanism
European Patent Office | 2024
|Adaptive traffic signal control method based on reinforcement learning and self-attention mechanism
European Patent Office | 2024
|Deep Reinforcement Learning-Based Traffic Signal Control
ASCE | 2024
|Deep Reinforcement Learning-based Traffic Signal Control
IEEE | 2023
|