The invention discloses a traffic signal lamp control method based on migration reinforcement learning, and relates to the technical field of intelligent traffic control, a model-based pre-training algorithm designed by the invention can predict the observation of a next state, and provides a pre-training algorithm for realizing cooperation and knowledge sharing among multiple intersections. A graph convolutional network is used to obtain feature information of a current intersection and neighbor intersections thereof, and a multi-head attention mechanism is used to determine influence degrees of different intersections on the current intersection, so that features and rules of traffic flow are better reflected. When a new intersection is trained, the source strategy trained in the pre-training stage is utilized to accelerate the training process, the source strategy most suitable for the current environment is selected in a probability distribution mode, the optimal source strategy can be found in a short time, the data efficiency is greatly improved, and the training cost is reduced. The problems of low data efficiency, high training cost and the like of some current traffic signal control algorithms based on deep strong learning are effectively solved.

    本发明公开了一种基于迁移强化学习的交通信号灯控制方法,涉及智能交通控制技术领域,本发明设计的基于模型的预训练算法,可以对下一状态的观测进行预测,为实现多交叉口之间的协作和知识共享,使用图卷积网络来获取当前路口及其邻居路口的特征信息,并且使用多头注意力机制来确定不同路口对当前路口的影响程度,从而更好地反映交通流的特征和规律。当训练新的路口时,利用预训练阶段训练的源策略来加速训练过程,使用概率分布的方式来选择最适合当前环境的源策略,能够在短时间内找到最佳的源策略,大大提高了数据效率,并降低了训练成本。本发明有效解决了当前一些基于深度强学习的交通信号控制算法存在的数据效率低、训练成本高等问题。


    Access

    Download


    Export, share and cite



    Title :

    Traffic signal lamp control method based on migration reinforcement learning


    Additional title:

    一种基于迁移强化学习的交通信号灯控制方法


    Contributors:
    ZHANG CHENGWEI (author) / LIU XIN (author) / LI YIHONG (author)

    Publication date :

    2023-09-15


    Type of media :

    Patent


    Type of material :

    Electronic Resource


    Language :

    Chinese


    Classification :

    IPC:    G06N COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS , Rechnersysteme, basierend auf spezifischen Rechenmodellen / G08G Anlagen zur Steuerung, Regelung oder Überwachung des Verkehrs , TRAFFIC CONTROL SYSTEMS



    Crossroad traffic signal lamp control method based on reinforcement learning

    WU LIBING / WANG MIN / ZHANG RUI et al. | European Patent Office | 2020

    Free access

    Pavement traffic signal lamp coordination control method based on reinforcement learning

    ZHU FEI / ZHU HAIJUN / FU YUCHEN et al. | European Patent Office | 2015

    Free access

    Traffic signal lamp control method and system based on reinforcement learning

    LUO JUAN / ZHENG YANLIU | European Patent Office | 2021

    Free access

    Traffic signal lamp control system and method based on reinforcement learning and dynamic timing

    LIU BINGYI / SONG BIN / WU PENGCHUN et al. | European Patent Office | 2021

    Free access

    Deep Reinforcement Learning-based Traffic Signal Control

    Ruan, Junyun / Tang, Jinzhuo / Gao, Ge et al. | IEEE | 2023