1 School of Traffic and Transportation, Lanzhou Jiaotong University, Lanzhou 730070, China
2 Construction Business Department, State Grid Gansu Electric Power Company, Lanzhou 730000, China
3 Digital Business Department, State Grid Gansu Electric Power Company, Lanzhou 730046, China
Show less
文章历史+
Received
Published
Revised
2025-05-23
2026-05-28
2025-12-16
Online Date
2026-08-07
PDF (9089K)
摘要
针对智能网联混合交通环境下车辆轨迹调控与信号配时协同优化难题,提出一种基于深度强化学习的双模块协同控制框架。通过构建时距管理-信号配时双环耦合机制实现动态优化:时距管理模块采用设计具有动态感知信号相位能力改进CACC(cooperative adaptive cruise control)跟驰模型,并通过时空约束生成CAVs(connected and automated vehicles)速度引导策略,确保车流连续平稳运行;信号配时模块设计具有动态感知信号相位能力的D3QN(double dueling deep Q-network)算法,构建包含相位长度、累计排队长度等34维特征的状态空间,以车均排队时间作为目标奖励函数。采用双网络架构与优先经验回放机制,通过ε-衰减策略平衡探索与开发,实现信号相位时长动态决策。通过SUMO-Python联合仿真平台进行多场景验证实验,在兰州市东方红广场西侧交叉口交通流量条件下进行多场景验证实验。对比实验表明:相较于传统固定配时方法、基准D3QN算法及传统车路协同算法,本模型使平均排队长度、平均车速和排队时间等指标均有所优化。研究成果验证了车辆轨迹控制与信号配时优化的深度耦合对交叉口通行效率的提升作用,为智能网联混合交通流场景下的车辆轨迹—信号配时协同控制提供具备研究价值的解决方案。
Abstract
To address the cooperative optimization challenge of vehicle trajectory regulation and signal timing in intelligent and connected hybrid traffic environments, a dual-module cooperative control framework based on deep reinforcement learning was proposed. A dual-loop coupling mechanism integrating time-space management and signal timing was developed for dynamic optimization. The time-space management module adopted an improved CACC (cooperative adaptive cruise control) car-following model with dynamic signal phase awareness, generating speed guidance strategies for connected and automated vehicles (CAVs) through spatiotemporal constraints to ensure continuous traffic flow. The signal timing module designed a double dueling deep Q-network (D3QN) algorithm with dynamic phase perception capability, constructing a 34-dimensional state space containing phase duration, cumulative queue length, and other parameters. A reward function targeting average vehicle queuing time was established. The algorithm utilized dual-network architecture and prioritized experience replay mechanism, achieving dynamic phase duration decisions through an ε-decay strategy to balance exploration and exploitation. Multi-scenario verification experiments were conducted using the SUMO-Python co-simulation platform under traffic flow conditions at the western intersection of Dongfanghong Square in Lanzhou City. Comparative tests demonstrated that compared with traditional fixed-time control, baseline D3QN algorithm, and conventional vehicle-infrastructure cooperative algorithms, the proposed model improved key metrics including average queue length, average speed, and queuing time. The results validate that deep coupling between vehicle trajectory control and signal timing optimization enhances intersection efficiency. The study provides an valuable research solution for cooperative vehicle trajectory-signal timing control in intelligent and connected hybrid traffic scenarios.
因此,研究开始进一步转向信号-轨迹协同优化的思路,旨在通过统一建模框架同时刻画信号控制策略与车辆行为决策,使两者在时空层面形成闭环反馈。Li等[19]在理想自动驾驶场景下构建信号—轨迹联合优化模型,通过枚举算法获得最优周期与绿灯相位;Yu等[20]采用混合整数线性规划(mixed-integer linear programming,MILP)方法,综合考虑车辆转向、信号相位顺序与持续时间等约束,建立了分层多目标优化框架;Guo等[21]提出结合动态规划与启发式算法(dynamic programming-simple heuristic,DP-SH)的求解策略,在降低计算时间的同时保持较优解质量;Tajalli等[22]通过线性化非线性约束将问题转化为可求解的混合整数线性规划,并利用松弛分解实现车道级控制优化;Lin等[23]则从空间结构划分的角度出发,将交叉口划分为核心区、缓冲区与自由区,分别对信号与轨迹进行分段优化。这些研究推动了从分层独立控制向跨层协同优化的转变,但其优化求解依赖显式交通模型,在复杂交通扰动与异质车流环境下仍面临泛化与鲁棒性不足的问题。
车头时距(time headway, THW)作为前后车辆时间间隔的量化表征,是影响交通流安全性、通行效率及能耗特性的核心参数。对于CAVs,现有研究主要通过美国加州大学PATH(Program for Appropriate Technology in Health)实验室提出的协同自适应巡航控制(CACC)模型实现车头时距的动态调控,但文献[24]研究表明,当期望车头时距参数设定失当时,模型会产生非物理性车辆行为(如负间距或倒车现象),严重制约了其在复杂交通场景中的应用可靠性;并且针对这个问题,基于PATH实验室模型框架,提出了一种融合时变期望间距的改进型车头时距优化模型,期望时距由当前车辆速度决定,以确定其跟车间距。时变期望间距的计算公式为
MikamiS, KakazuY. Genetic reinforcement learning for cooperative traffic signal control[C]// The First IEEE Conference on Evolutionary Computation. Orlando: IEEE, 1994: 223-228.
[2]
EltantawyS, AbdulhaiB, AbdelgawadH. Design of reinforcement learning parameters for seamless application of adaptive traffic signal control[J]. Journal of Intelligent Transportation Systems, 2014, 18(3): 227-245.
[3]
LiL, LüY, WangF Y. Traffic signal timing via deep reinforcement learning[J]. IEEE/CAA Journal of Automatica Sinica, 2016, 3(3): 247-254.
RenAnni, ZhouDake, FengJinhao, et al. Attention mechanism based deep reinforcement learning for traffic signal control[J]. Application Research of Computers, 2023, 40(2): 430-434.
[6]
Van HasseltH. Double Q-learning[C]// Advances in Neural Information Processing Systems 29 (NeurIPS 2016). Barcelona: Neural Information Processing Systems Foundation, 2016: 2613-2621.
SunHao, ChenChunlin, LiuQiong, et al. Traffic signal control method based on deep reinforcement learning[J]. Computer Science, 2020, 47(2): 169-174.
[9]
HesselM, SoyerH, SchaulT, et al. Learning to play in a day: faster deep reinforcement learning by optimal transport[C]// Proceedings of the 35th International Conference on Machine Learning (ICML 2018). Stockholm: PMLR, 2018: 1694-1703.
SongTailong, HeYulong, LiuQin. Signal control of key intersections in large-scale events based on deep reinforcement learning[J]. Science Technology and Engineering, 2023, 23(22): 9694-9701.
SunWei, ZhangMengya, MaChengyuan, et al. Coordination of signal and vehicle trajectory at intersections for mixed traffic flow[J]. Journal of Transportation Systems Engineering and Information Technology, 2023, 23(1): 97-105.
LiZhenlong, ZhangJingsi, LiuQin, et al. Coordinated control method of double-cycling arterial based on improved Q-learning[J]. Science Technology and Engineering, 2021, 21(29): 12744-12750.
LiuZhimin, YeBaolin, ZhuYaodong, et al. Traffic signal control method based on deep reinforcement learning[J]. Journal of Zhejiang University (Engineering Science), 2022, 56(6): 1249-1256.
TangMuyao, ZhouDake, LiTao. State prediction based on deep reinforcement learning for traffic signal control[J]. Application Research of Computers, 2022, 39(8): 2311-2315.
[24]
RakhaH, KamalanathsharmaR K. Eco-driving at signalized intersections using V2I communication[C] // The 14th International IEEE Conference on Intelligent Transportation Systems. New York: IEEE, 2011: 341-346.
[25]
AsadiB, VahidiA. Predictive cruise control: utilizing upcoming traffic signal information for improving fuel economy and reducing trip time[J]. IEEE Transactions on Control Systems Technology: A Publication of the IEEE Control Systems Society, 2011, 19(3): 707-714.
[26]
NgoduyD. Platoon-based macroscopic model for intelligent traffic flow[J]. Transportmetrica B: Transport Dynamics, 2013, 1(2): 153-169.
[27]
NgoduyD, JiaD. Multi anticipative bidirectional macroscopic traffic model considering cooperative driving strategy[J]. Transportmetrica B: Transport Dynamics, 2017, 5(1): 96-110.
[28]
LiZ, ElefteriadouL, RankaS. Signal control optimization for automated vehicles at isolated signalized intersections[J]. Transportation Research Part C: Emerging Technologies, 2014, 49: 1-18.
[29]
YuC, FengY, LiuH X, et al. Integrated optimization of traffic signals and vehicle trajectories at isolated urban intersections[J]. Transportation Research Part B: Methodological, 2018, 112: 89-112.
[30]
GuoY, MaJ Q, XiongC F, et al. Joint optimization of vehicle trajectories and intersection controllers with connected automated vehicles: combined dynamic programming and shooting heuristic approach[J]. Transportation Research Part C: Emerging Technologies, 2019, 98: 54-72.
[31]
TajalliM, HajbabaieA. Traffic signal timing and trajectory optimization in a mixed autonomy traffic stream[J]. IEEE Transactions on Intelligent Transportation Systems, 2021, 23(7): 6525-6538.
[32]
LinP, LiuJ, JinP J, et al. Autonomous vehicle-intersection coordination method in a connected vehicle environment[J]. IEEE Intelligent Transportation Systems Magazine, 2017, 9(4): 37-47.
[33]
ZhaoH, LiuS, LiangR, et al. ACC/CACC vehicle car-following model based on a time-varying expected spacing and a stability analysis of mixed traffic flow[J]. Journal of Shanghai Jiaotong University (Science), 2025, 30(2): 1-13.