The invention discloses a multi-intersection traffic signal control method based on deep reinforcement learning, and the method comprises the following two parts: obtaining traffic information based on deep learning target detection: obtaining the traffic information of each road of each intersection according to an intersection camera, and carrying out the detection based on deep learning; performing signal lamp phase control based on a deep reinforcement learning DQN algorithm: performing intelligent decision making in combination with the traffic information identified by the target detection model and the deep reinforcement learning DQN algorithm, and selecting a signal lamp phase control strategy most beneficial to road congestion condition alleviation by the Deep-QNet algorithm after obtaining road condition information through deep learning target detection; according to the method, a complex and changeable road environment can be processed, the feature extraction capability of the neural network is more effective along with the increase of intersections included in the system, and a better effect is achieved in signal lamp overall decision making for environmental traffic pressure.
本发明公开了基于深度强化学习的多交叉路口交通信号控制方法,包含以下两个部分:基于深度学习目标检测的交通信息获取:根据十字路口摄像头获取每个路口各条道路的交通信息,并基于深度学习进行检测;基于深度强化学习DQN算法的信号灯相位控制:结合目标检测模型识别到的交通信息与深度强化学习DQN算法进行智能决策,在通过深度学习目标检测获取道路路况信息后,Deep‑QNet算法会选择最有利于道路拥堵情况缓解的信号灯相位控制策略;本方法能处理复杂多变的道路环境,随着系统包含的路口增加,神经网络特征提取能力会更加富有成效,使得在对于环境交通压力的信号灯统筹决策中发挥更好的效果。
Multi-intersection traffic signal control method based on deep reinforcement learning
基于深度强化学习的多交叉路口交通信号控制方法
2024-01-16
Patent
Elektronische Ressource
Chinesisch
Multi-intersection traffic signal control method based on deep reinforcement learning
Europäisches Patentamt | 2023
|Single-intersection signal control method based on deep reinforcement learning algorithm
Europäisches Patentamt | 2023
|Deep Reinforcement Learning-based Traffic Signal Control
IEEE | 2023
|Multi-intersection graph deep reinforcement learning signal lamp control method under data loss
Europäisches Patentamt | 2024
|