The invention provides a self-adaptive traffic signal control method based on graph deep reinforcement learning. The existing traffic signal control method based on deep reinforcement learning mostly uses a standard neural network, such as a convolutional neural network, to perceive a traffic state, but the methods often ignore the influence of internal relation characteristics of the traffic state on control selection. Therefore, according to the method, the graph neural network and reinforcement learning are fused, a graph structure data representation form based on multi-dimensional feature fusion is adopted, the potential correlation information of the traffic state is mined by utilizing the efficient extraction capability of the graph neural network on nonlinear depth features, and the adaptive strategy is fed back in combination with the deep reinforcement learning; effective acquisition of traffic information and adaptive control of scheduling are realized. Compared with a traditional traffic signal control method based on deep reinforcement learning, the method better improves the effectiveness of decision information and the robustness of a scheduling method in traffic signal control.
本发明提出一种基于图深度强化学习的自适应交通信号控制方法。现有的基于深度强化学习的交通信号控制方法大多数使用标准神经网络,例如卷积神经网络来感知交通状态,但这些方法往往忽略了交通状态的内在关系特征对控制选择的影响。为此,本发明将图神经网络和强化学习相融合,采用基于多维度特征融合的图结构数据表示形式,利用图神经网络对非线性深度特征的高效提取能力,挖掘交通状态的潜在关联信息,结合深度强化学习反馈适配策略,实现交通信息的有效获取与调度的自适应控制。相较于传统深度强化学习的交通信号控制方法,本发明在交通信号控制中更好的提升了决策信息的有效性与调度方法的鲁棒性。
Self-adaptive traffic signal control method based on graph deep reinforcement learning
一种基于图深度强化学习的自适应交通信号控制方法
2022-06-14
Patent
Elektronische Ressource
Chinesisch
Deep Reinforcement Learning-based Traffic Signal Control
IEEE | 2023
|Traffic signal control method based on deep reinforcement learning
Europäisches Patentamt | 2021
|Adaptive Traffic Signal Control Model on Intersections Based on Deep Reinforcement Learning
DOAJ | 2020
|