摘要
信控交叉口车辆频繁启停是道路交通高能耗的关键场景之一。面向信控交叉口的生态驾驶,可基于实时信号灯信息,控制车辆不停车通过交叉口,从而提高车辆生态性、降低道路交通能耗。然而,当前信控交叉口生态驾驶方法主要关注固定信号交叉口,鲜有面向感应信号交叉口的研究。为解决感应信号配时方案动态变化所造成的不确定性挑战,提出了基于数据驱动模型预测控制的感应信号交叉口生态驾驶方法,主要包含以下3项创新贡献:①提出面向感应信号交叉口的生态驾驶策略,解决了感应信号交叉口环境下车辆最优生态速度规划难题;②采用异构图神经网络模型,建立感应信号配时动态预测方法;③实现基于空间域的速度规划建模,解决了传统时间域下车辆速度规划和空间域下到达时间约束之间建模域不匹配的问题。基于VISSIM仿真平台的测试验证结果显示:该方法能够准确预测感应信号方案,绿灯开始时间和绿灯结束时间的预测精度分别达到2.8 s和2.9 s,且能保证81%以上的预测结果误差小于3 s。所提出的生态驾驶速度规划方法可控制车辆不停车高效通过交叉口,平均节约油耗8.0%、驾驶安全性提升20%。模型平均计算时间小于40 ms,能够保证实际应用中的计算效率。
Frequent stop-and-go driving at signal-controlled intersections is one of the key scenarios contributing to high-energy consumption in road traffic.Eco-driving at signal-controlled intersections leverages real-time traffic signal information to enable vehicles to pass through intersections without stopping,thus enhancing vehicle eco-friendliness and reducing road trafficenergy consumption.However,current eco-driving methods primarily focus on fixed-time signalintersections.Limited research has been conducted to address actuated signal intersections.Toaddress the uncertainty challenges posed by the dynamic changes in actuated signal timing,thispaper proposes a data-driven model,predictive control-based,eco-driving method for actuatedsignal intersections according to the following three innovative contributions:① Proposition of aneco-driving strategy for actuated signal intersections,solving the challenge of optimal eco-speedplanning for vehicles in actuated signal intersections;② adoption of a heterogeneous graph neuralnetwork model was adopted to establish a dynamic prediction method for actuated signal timing,and ③ a spatial-domain-based,speed planning model was developed,resolving the mismatchbetween traditional temporal-domain speed planning and spatial-domain arrival time constraints.Validation results based on the VISSIM simulation platform show that the proposed method canaccurately predict actuated signal timing with prediction accuracies for green light start and endtimes reaching 2.8 s and 2.9 s,respectively,and >81% of the prediction results exhibit errors> 3 s. The proposed eco-driving speed planning method allows vehicles to pass throughintersections efficiently without stopping,achieving an average fuel saving of 8.0% and drivingsafety improvements of 20%.The model's average computation time is less than 40 ms,ensuringcomputational efficiency in practical applications.
作者
冯永威
李朔远
胡笳
王浩然
FENG Yong-wei;LI Shuo-yuan;HU Jia;WANG Hao-ran(Key Laboratory of Road and Traffic Engineering,Ministry of Education,Tongji University,Shanghai 201804,China)
出处
《中国公路学报》
北大核心
2025年第8期57-69,共13页
China Journal of Highway and Transport
基金
新一代人工智能国家科技重大专项(2022ZD0115503)
国家重点研发计划项目(2022YFF0604905)
国家自然科学基金项目(52302412)
长三角科技创新共同体联合攻关项目(2023CSJGG0800)
2023年度国家资助博士后研究人员计划项目(GZB20230519)
上海启明星培育项目扬帆计划项目(23YF1449600)
上海市超级博士后计划项目(2022571)
中国博士后科学基金项目(2022M722405)。
关键词
交通工程
生态驾驶
模型预测控制
智能网联车辆
图神经网络
traffic engineering
eco-driving
model predictive control
connected and automated vehicle
graph neural network
