考虑换道风险动态评估的快速路合流区CAV换道决策模型  

CAV Lane Change Decision-making Model in Urban Expressway Merging Areas Considering Dynamic Lane Change Risk Assessment

作　　者：程国柱 孟凤威 陈永胜 吕家乐 王文志 CHENG Guo-zhu;MENG Feng-wei;CHEN Yong-sheng;LYU Jia-le;WANG Wen-zhi(School of Civil Engineering and Transportation,Northeast Forestry University,Harbin 150040,Heilongjiang,China)

机构地区：[1]东北林业大学土木与交通学院,黑龙江哈尔滨150040

出　　处：《中国公路学报》2025年第1期268-280,共13页China Journal of Highway and Transport

基　　金：中央高校基本科研业务费专项资金项目(2572023CT21);国家自然科学基金项目(52378433);黑龙江省重点研发计划项目(JD22A014)。

摘　　要：城市快速路作为重要交通走廊,其合流区交通调度更为复杂。为保障智能网联车辆(Connected and Automated Vehicle, CAV)在城市快速路合流区高效、安全、舒适行驶,基于城市快速路合流区的换道特征,开展了考虑换道风险动态评估结果的CAV换道决策模型的相关研究。首先,对换道车辆与邻车进行时空重叠分析,将分析结果作为潜在风险判别依据,对存在时空重叠的CAV及其邻车计算换道碰撞时间(Lane Change Time to Collision, LCTTC),实现换道风险动态评估;然后,将换道风险动态评估结果融入多目标奖励函数优化的DQN(Deep Q-Network)网络结构,并综合考虑车辆行驶高效性、安全性与舒适性等因素,提出具有风险感知的SIDQN(Security Improvement Deep Q-Network)换道决策;最后,通过仿真试验进行验证。研究结果表明:相较于对比策略,提出的SIDQN策略换道成功率保持在95%以上,且运行平均速度不低于21.008 m·s^(-1);此外,在复杂的交通合流场景中,融入安全性奖励的SIDQN策略表现出最佳安全性,其LCTTC平均占比仅为9.56%,远低于其他对比策略,同时事故率统计结果持续保持较低水平;在舒适性方面,SIDQN策略下换道次数仅为4次,连续换道和无效换道次数均为1次,并显著降低了频繁换道和极端加速减速操作次数,提升了乘客的舒适体验。提出的换道决策模型综合性能优势明显,可为智能网联环境下城市快速路合流区CAV安全换道决策研究提供参考。Urban expressways,as essential transportation corridors,present unique challenges for traffic management,especially in merging zones.This study proposes a lane change decision-making model for connected and automated vehicles(CAVs)designed to optimize their operation in these complex environments.Specifically,the model addresses the unique lane change dynamics in urban expressway merging zones,focusing on safety,efficiency,and comfort.The analysis begins with a spatiotemporal overlap evaluation between lane-changing vehicles and adjacent traffic,providing the basis for identifying potential collision risks.For CAVs and neighboring vehicles exhibiting spatiotemporal overlap,the lane change time to collision(LCTTC)is computed,enabling dynamic risk assessment.The resulting risk metrics are integrated into a multi-objective reward function to optimize the deep q-network(DQN)architecture,which balances vehicle safety,operational efficiency,and passenger comfort.A novel,risk-aware lane change strategy,termed the security improvement deep q-network(SIDQN),is then proposed.Simulation experiments validate the effectiveness of this strategy,with results demonstrating a lane change success rate exceeding 95%and an average speed of no less than 21.008 m·s^(-1).Moreover,the SIDQN strategy improves safety performance in complex merging scenarios,reducing the LCTTC ratio to just 9.56%,a substantial decrease compared to baseline strategies.The accident rate remains minimal.Additionally,the SIDQN strategy limits the number of lane changes to four and minimizes ineffective consecutive lane changes,reducing extreme acceleration and deceleration events and thereby enhancing passenger comfort.In conclusion,the proposed lane change decision-making model significantly improves performance in urban expressway merging zones and provides a valuable reference for advancing CAV safety and comfort in intelligent,connected environments.

关 键 词：交通工程 换道决策 深度强化学习 智能网联车辆 换道风险评估 快速路合流区 

分 类 号：U491[交通运输工程—交通运输规划与管理]