基于CBAM注意力机制的智能交通信号控制
Intelligent Traffic Control Technology Based on CBAM Attention Mechanism
于贺婷 1刘思萌 1文峰1
作者信息
- 1. 沈阳理工大学 信息科学与工程学院,沈阳 110159
- 折叠
摘要
针对智能交通系统存在的卷积神经网络特征提取能力弱和特征表达能力有待提升等问题,在深度双Q网络(double deep Q network,Double DQN)模型基础上提出一种基于卷积注意力模块(convolutional block attention module,CBAM)的深度强化学习模型,用于智能交通信号控制.在三维卷积神经网络中加入CBAM轻量注意力模块,通过通道注意力和空间注意力两个模块结构更好地捕捉特征之间的相互依赖关系,增强卷积神经网络的特征表示质量,从而提升对拥堵路段重点特征的关注度以缓解交通拥堵问题.在城市交通仿真器SUMO(simula-tion of urban mobility)上的实验结果表明,相较其他常用算法,本文算法提高了交通灯配时的效率和稳定性,可为交通配时优化技术提供可靠依据.
Abstract
A deep reinforcement learning model is proposed based on the double deep Q network(Double DQN)model,using the convolutional block attention module(CBAM)to address issues such as weak feature extraction capabilities and limited feature expression in convolutional neural networks within intelligent transportation systems.By integrating the lightweight CBAM attention module into the 3D convolutional neural network,the model can better capture the interdependen-cies between features through the channel attention and spatial attention modules.This enhances the quality of feature representation in the convolutional neural network,thereby improving the focus on key features of congested road sections and alleviating traffic congestion.Experimental results con-ducted on the SUMO(simulation of urban mobility)urban traffic simulator demonstrate that the proposed algorithm improves the efficiency and stability of traffic signal timing compared to other commonly used algorithms,providing a reliable basis for traffic timing optimization technology.
关键词
交通信号控制/深度强化学习/深度双Q网络/卷积注意力模块Key words
traffic signal control/deep reinforcement learning/deep double Q network/convolu-tional block attention module引用本文复制引用
基金项目
国家重点研发计划"社会治理与智慧社会科技支撑"重点专项(2022YFC3302502)
出版年
2024
NETL
NSTL
万方数据