The invention discloses a multi-intersection graph deep reinforcement learning signal lamp control method under data loss, and the method comprises the steps: firstly building a multi-intersection signal lamp control scene based on an SUMO simulation platform, and defining the state space of an intelligent agent; secondly, repairing an incomplete agent state space based on a generative adversarial network (WGAN); thirdly, constructing a graph neural network to perform effective interaction and characterization on agent information, constructing a dynamic interaction graph among multiple agents based on a two-stage attention network, aggregating state features of the multiple agents based on GraphSage, and outputting a Q value based on a mapping decision Q value network; then, constructing a phase action space of the intelligent agent, and defining a reward function to guide the intelligent agent to make an optimal decision; and finally training and testing the model. According to the invention, high multi-intersection signal lamp control efficiency can be realized in the absence of traffic states.
一种数据缺失下的多交叉口图深度强化学习信号灯控制方法,首先搭建基于SUMO仿真平台的多交叉口信号灯控制场景,定义智能体的状态空间;其次,基于生成对抗网络(WGAN)对不完整的智能体状态空间进行修复;再次,构建图神经网络对智能体信息进行有效交互和表征,基于两阶段注意力网络构建多智能体间动态交互图,基于GraphSage聚合多智能体状态特征,基于映射决策Q值网络输出Q值;然后,构建智能体的相位动作空间,定义奖励函数来引导智能体做出最佳决策;最后对模型进行训练和测试。本发明能够在交通状态缺失下实现较高的多交叉口信号灯控制效率。
Multi-intersection graph deep reinforcement learning signal lamp control method under data loss
数据缺失下的多交叉口图深度强化学习信号灯控制方法
2024-03-15
Patent
Elektronische Ressource
Chinesisch
IPC: | G08G Anlagen zur Steuerung, Regelung oder Überwachung des Verkehrs , TRAFFIC CONTROL SYSTEMS / G06F ELECTRIC DIGITAL DATA PROCESSING , Elektrische digitale Datenverarbeitung / G06N COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS , Rechnersysteme, basierend auf spezifischen Rechenmodellen |
Signal lamp-free intersection control method and device based on deep reinforcement learning
Europäisches Patentamt | 2023
|Multi-intersection traffic signal control method based on deep reinforcement learning
Europäisches Patentamt | 2024
|Multi-intersection traffic signal control method based on deep reinforcement learning
Europäisches Patentamt | 2023
|