融合双注意力机制的快速路协同深度强化学习方法

doi:10.16097/j.cnki.1009-6744.2026.02.013

摘要/Abstract

摘要： 城市快速路匝道合流区是交通瓶颈频发区域，而网联自动驾驶车辆与人工驾驶车辆构成的混合交通流为交通控制带来新挑战。本文旨在提出一种兼顾匝道排队溢出影响的混合交通流可变限速控制方法，以缓解拥堵并增强交通流稳定性。研究将合流区协同优化控制问题抽象为马尔可夫决策过程，提出双深度Q网络和双注意力机制（Convolutional Block Attention Module, CBAM）的集成方法DDQN-CBAM：构建含匝道排队长度和合流区密度等多维度核心参数与网格化特征的扩展状态空间，引入CBAM双注意力机制强化关键特征提取，设计融合通行效率和排队控制等多目标的奖励函数，并结合优先经验回放和渐进式车流输入等策略优化训练。以徐州北三环快速路为实例，在仿真平台（SUMO）进行仿真验证。实验结果表明，该方法较传统控制策略，总行程时间减少26.49%，总行驶距离提升35.95%，交通流量标准差降低超22.5%，小时控制频率与速度调整率分别稳定在10次和0.14左右。本文提出的方法兼具工程适用性与鲁棒性，为城市快速路匝道合流区交通控制提供可靠支撑。

关键词: 智能交通, 可变限速控制, 深度强化学习, 城市快速路, 合流区协同影响, 注意力机制

Abstract: On urban expressways, ramp merging areas are prone to become traffic bottlenecks. The mixed traffic flow composed of connected and automated vehicles (CAVs) and human-driven vehicles (HDVs) brings new challenges to traffic control. This study proposes a variable speed limit (VSL) control method for mixed traffic flow that considers ramp queuing spillover impacts, so as to alleviate congestion and enhance traffic flow stability. The merge area collaborative optimization control problem is formulated as a Markov decision process (MDP), and an integrated method DDQN-CBAM is proposed by combining the double deep Q-network (DDQN) and convolutional block attention module (CBAM). Specifically, an extended state space is constructed, including multi dimensional core parameters such as ramp queue length and merging area density, as well as grid-based features. The CBAM dual attention mechanism is introduced to strengthen the extraction of key features. A reward function integrating multi-objectives such as traffic efficiency and queue control is designed, and the training process is optimized by combining strategies such as prioritized experience replay and progressive traffic input. Taking the North Third Ring Road Expressway of Xuzhou, China as a case study, validation is completed on the Simulation of Urban Mobility (SUMO) platform. Experimental results show that compared with traditional control strategies, the proposed method reduces the total travel time by 26.49%, increases the total travel distance by 35.95%, decreases the standard deviation of traffic flow by more than 22.5%, and stabilizes the hourly control frequency and speed adjustment rate at approximately 10 times and 0.14, respectively. This method possesses both engineering applicability and robustness, and provides reliable support for traffic control of ramp merging areas on urban expressways.

Key words: intelligent transportation, variable speed limit control, deep reinforcement learning, urban expressway, merge area collaborative impacts, attention mechanism

中图分类号:

U495

孙健, 纪裕伟, 于珂伟, 李子豪, 赵昱霖. 融合双注意力机制的快速路协同深度强化学习方法[J]. 交通运输系统工程与信息, 2026, 26(2): 137-147.

SUN Jian, JI Yuwei, YU Kewei, LI Zihao, ZHAO Yulin. Collaborative Deep Reinforcement Learning Method for Expressways Integrating Dual Attention Mechanism[J]. Journal of Transportation Systems Engineering and Information Technology, 2026, 26(2): 137-147.

导出引用管理器 EndNote|Ris|BibTeX

链接本文: http://www.tseit.org.cn/CN/10.16097/j.cnki.1009-6744.2026.02.013

http://www.tseit.org.cn/CN/Y2026/V26/I2/137

参考文献

[1]孙静怡,沈俊江,刘拥华,等.城市快速路可变限速策略[J]. 公路交通科技,2012, 29(11): 98-103. [SUN J Y, SHEN J J, LIU Y H, et al. Variable speed limits strategy of urban expressway[J]. Journal of Highway and Transportation Research and Development, 2012, 29 (11): 98-103.]

[2]田江磊,李博,彭瑾晗,等.高速公路主线拥堵区域可变限速控制方法[J].公路,2025,70(8): 282-293. [TIAN J L, LI B, PENG J H, et al. Variable speed limit control method for congested areas on freeway mainline[J]. Highway, 2025, 70(8): 282-293.]

[3] 朱健,丁鹏程.非常发性瓶颈区可变限速控制方法[J]. 交通科技与经济, 2020, 22(1): 38-43, 58. [ZHU J, DING P C. Variable speed limit control method fornon recurrent bottleneck zone[J]. Technology and Economy in Areas of Communications, 2020, 22(1): 38-43, 58.]

[4] 邓炜,于欣海,张朗,等.基于5G边缘计算的高速公路可变限速与换道协同控制[J].合肥工业大学学报（自然科学版）,2025, 48(5): 642-650. [DENG W, YU X H, ZHANG L, et al. Collaborative control of variable speed limits and lane change on freeway based on 5G edge computing[J]. Journal of Hefei University of Technology (Natural Science), 2025, 48(5): 642-650.]

[5]李彬,牛万龙,胡纪龄,等.基于大数据分析的高速公路分车型拥堵态势评估[J].公路交通科技,2024,41 (8): 40-54. [LI B, NIU W L, HU J L, et al. Expressway congestion state assessment due to different vehicle types based on big data[J]. Journal of Highway and Transportation Research and Development, 2024, 41(8): 40-54.]

[6]余荣杰,徐灵,章锐辞.基于多智能体深度强化学习的高速公路可变限速协同控制方法[J].同济大学学报(自然科学版),2024, 52(7): 1089-1098. [YU R J, XU L, ZHANG R C. Coordinated variable speed limit control for freeway based on multi-agent deep reinforcement learning[J]. Journal of Tongji University (Natural Science), 2024, 52(7): 1089-1098.]

[7]韩磊,张轮,郭为安.混合交通流环境下基于改进强化学习的可变限速控制策略[J].交通运输系统工程与信息, 2023, 23(3): 110-122. [HAN L, ZHANG L, GUO W A. Variable speed limit control based on improved dueling double deep Q network under mixed traffic environment[J]. Journal of Transportation Systems Engineering and Information Technology, 2023, 23(3): 110-122.]

[8] 王玥骄,卢霄娟,郑诗禹,等.智能网联环境下高速公路事故临近区可变限速控制[J].长安大学学报(自然科学版),2025, 45(3): 152-162. [WANG Y J, LU X J, ZHENG S Y, et al. Variable speed limit control for expressway accident vicinity area in intelligent connected environment[J]. Journal of Chang'an University (Natural Science Edition), 2025, 45(3): 152-162.]

[9] 白如玉,焦朋朋,陈越,等.基于强化学习的车道级可变限速控制策略[J]. 交通信息与安全, 2024, 42(1): 105-114. [BAI R Y, JIAO P P, CHEN Y, et al. Differential variable speed limit control strategy based on reinforcement learning[J]. Journal of Transport Information and Safety, 2024, 42(1): 105-114.]

[10] 蔡田茂, 孔伟伟,罗禹贡,等.基于MADDPG算法的匝道合流区多车协同控制[J]. 汽车安全与节能学报, 2024, 15(6): 923-933. [CAI T M, KONG W W, LUO Y G, et al. Multi-vehicle cooperative control in ramp merging area based on MADDPG algorithm[J]. Journal of Automotive Safety and Engergy, 2024, 15(6): 923-933.]

[11] 尚春琳, 刘小明,田玉林,等.基于深度强化学习的综合干线协调控制方法[J].交通运输系统工程与信息, 2021, 21(3): 64-70. [SHANG C L, LIU X M, TIAN Y L, et al. Priority of dedicated bus arterial control based on deep reinforcement learning[J]. Journal of Transportation Systems Engineering and Information Technology, 2021, 21(3): 64-70.]

[12] 周浩, 胡坚明,张毅,等.快速路可变限速与匝道控制协同优化策略[J].交通运输系统工程与信息,2017,17 (2): 68-75. [ZHOU H, HU J M, ZHANG Y, et al. A coordinated optimization strategy of variable speed limit and ramp metering for expressway[J]. Journal of Transportation Systems Engineering and Information Technology, 2017, 17(2): 68-75.]

[13] 邬岚, 任斯奇,陈茜,等.基于车路协同的快速路合流区可变限速控制方法[J].重庆交通大学学报(自然科学版), 2024, 43(9): 68-77. [WU L, REN S Q, CHEN Q, et al. Variable speed limit control method based on cooperative vehicle infrastructure system[J]. Journal of Chongqing Jiaotong University (Natural science), 2024, 43(9): 68-77.]

[14] 宿永辉, 欧阳涛,潘新福,等.面向高速公路连续瓶颈的协同可变限速控制[J].交通运输工程与信息学报, 2024, 22(3): 166-180. [SU Y H, OUYANG T, PAN X F, et al. A collaborative variable speed-limit control for continuous bottlenecks on freeways[J]. Journal of Transportation Engineering and Information, 2024, 22 (3): 166-180.]

[15] HE Z A, HAN Y, YU H, et al. Integrated feedback perimeter control-based ramp metering and variable speed Limits for multibottleneck freeways[J]. Journal of Transportation Engineering, Part A: Systems, 2024, 150 (9): 04024054.

[16] 席殊, 陈旭梅,李培坤,等.考虑CAV专用道汇入需求的高速公路交织区集成控制策略[J].哈尔滨工业大学学报,2025, 57(11): 1-11, 21. [XI S, CHEN X M, LI P K, et al. An integrated control strategy for freeway weaving area considering CAV dedicated lanemerging demand[J]. Journal of Harbin Institute of Technology, 2025, 57(11): 1-11, 21.]

[17] HE Z L, WANG L, SU Z C, et al. Integrating variable speed limit and ramp metering to enhance vehicle group safety and efficiency in a mixed traffic environment[J]. Physica A: Statistical Mechanics and its Applications, 2024, 641: 129754.

[18] XIE D F, ZHAO X M, HE Z B. Heterogeneous traffic mixing regular and connected vehicles: Modeling and stabilization[J]. IEEE Transactions on Intelligent Transportation Systems, 2019, 20 (6): 2060-2071.

[19] 孙健, 宋茂星,邱果,等.基于电动汽车大数据的多等级充电站选址与服务能力研究[J].中国公路学报, 2024, 37(4): 48-60. [SUN J, SONG M X, QIU G, et al. Location and service capability of multilevel charging stations based on electric vehicle big data[J]. China Journal of Highway and Transport, 2024, 37(4): 48-60.]

[20] MA M J, WU H D, S D J. Collaborative variable speed limit for urban expressway mainline and on-ramp in CAVs environment based on deep reinforcement learning [J]. Physica A: Statistical Mechanics and its Applications, 2026, 689: 131428.