The invention discloses a traffic signal control method based on reinforcement learning, and the method comprises the steps: observing a current road condition through a road network, generating a training sample, carrying out the training of a deep reinforcement learning (DQN) network which comprises a state space S, an action space A and an excitation function r, and finally generating a signal lamp prediction value through the trained DQN network, and carrying out the control of an actual traffic signal. According to the method, signal control is carried out by adopting a reinforcement learning method, so that the characteristic of real-time change of road conditions is fully considered, the throughput rate of a road network is improved, a traffic flow model is introduced, risk factors are strictly selected, and the problems caused by inherent defects of reinforcement learning are effectively avoided.
一种基于强化学习的交通信号控制方法,通过路网观测当前道路状况生成训练样本,对包含状态空间S、动作空间A和激励函数r的深度强化学习(DQN)网络进行训练,最终采用训练后的DQN网络生成信号灯预测值对实际交通信号进行控制。本发明采用强化学习的方法进行信号控制从而充分考虑路况的实时变化的特点,从而提升路网的吞吐率的同时,引入交通流模型,同时严格选取冒险因子,从而有效避免因为强化学习固有缺陷而带来的问题。
Traffic signal control method based on reinforcement learning
基于强化学习的交通信号控制方法
2023-02-17
Patent
Electronic Resource
Chinese
Deep Reinforcement Learning-based Traffic Signal Control
IEEE | 2023
|Traffic signal control method based on deep reinforcement learning
European Patent Office | 2021
|Crossroad traffic signal lamp control method based on reinforcement learning
European Patent Office | 2020
|A Method of Reinforcement Learning Based Automatic Traffic Signal Control
Tema Archive | 2011
|