交通运输系统工程与信息

交通运输系统工程与信息 ›› 2024, Vol. 24 ›› Issue (1): 66-80.DOI: 10.16097/j.cnki.1009-6744.2024.01.007

• 智能交通系统与信息技术 • 上一篇    下一篇

引力理论框架下基于综合竞争力的自动驾驶 拟人换道决策模型

裴玉龙*1，傅博涵1，王子奇1，张杰1, 2   

  1. 1. 东北林业大学，土木与交通学院，哈尔滨 150040；2. 宁德师范学院，信息与机电工程学院，福建 宁德 352100
  • 收稿日期:2023-11-07 修回日期:2023-12-08 接受日期:2023-12-20 出版日期:2024-02-25 发布日期:2024-02-11
  • 作者简介:裴玉龙(1961- )，男，黑龙江桦川人，教授，博士
  • 基金资助:
    国家自然科学基金 (50578056)

Comprehensive Competitiveness-based Autonomous Driving Human-imitative Lane-changing Model Under Gravity Theory

PEI Yulong*1, FU Bohan1, WANG Ziqi1, ZHANG Jie1, 2   

  1. 1. School of Civil and Traffic, Northeast Forestry University, Harbin 150040, China; 2. College of Information and Mechanical & Electrical Engineering, Ningde Normal University, Ningde 352100, Fujian, China
  • Received:2023-11-07 Revised:2023-12-08 Accepted:2023-12-20 Online:2024-02-25 Published:2024-02-11
  • Supported by:
    National Natural Science Foundation of China (50578056)

PDF

摘要： 为有效刻画城市快速路中自动驾驶环境下的车辆换道决策机理，考虑主车与周边车辆的位置属性、驾驶风格属性及车辆运动属性对主车换道行为的影响，建立基于综合竞争力的拟人换道决策模型。首先，以邻近前车间距、邻近前车速度差及驾驶风格这3种因素作为自动驾驶车辆的拟人化换道意愿属性，量化表征主车的换道意愿；然后，基于人类决策中的悲观主义准则，分析换道过程中周边车辆与主车可能产生的竞争行为，利用车头间距比和驾驶风格差异提出潜在竞争强度概念；其次，考虑环境稳定性对驾驶舒适性的影响，提出“速度伪距”“加速度伪距”概念，衡量换道后的环境稳定性；最后，结合引力理论建立以车辆横向速度为求解目标的综合竞争力换道决策模型。在模型标定中，筛选Ubiquitous Traffic Eyes开源数据集，得到非强制换道片段数据，利用蚁群算法标定模型参数。采用随机排列交叉验证方法进行验证，以正确率为模型精度和泛化能力的评价指标，并将其与传统模型进行对比。结果表明，将训练与验证比设置为72%∶28%、65%∶35%、57%∶43%和50%∶50%时，平均正确率区间为87.67%~90.34%，说明该模型具有较强的 鲁棒性和可行性，相比于传统模型，本文模型具有更高的预测精度，可为自动驾驶环境下车辆的车道选择提供依据。

关键词: 智能交通, 换道决策, 综合竞争力, 自动驾驶, 横向速度, 轨迹数据

Abstract: To effectively characterize the vehicle lane-changing decision-making mechanism in an automated driving environment on urban expressways, this paper proposes a human-imitative lane-changing decision model based on comprehensive competitiveness. The method considers the impact of the positional, driving style and vehicle motion attributes of the subject vehicle and the adjacent vehicles on the subject vehicle's lane-changing behavior. The three factors of adjacent front vehicle distance, speed difference and driving style were used as the human-imitative lane-changing willingness attributes of the autonomous vehicle to quantitatively characterize the subject vehicle's lane-changing willingness. Then, based on the pessimistic criterion in human decision-making, the potential competitive behavior between the adjacent vehicles and the subject vehicle in the process of changing lanes was analyzed. The concept of potential competitive intensity was proposed using the headway ratio and driving style differences. Considering the influence of environmental stability on driving comfort, this study uses the concepts of 'velocity pseudo-distance' and 'acceleration pseudo-distance' to measure the environmental stability after lane changing. A comprehensive competitiveness lane-changing decision model with vehicle lateral speed as the solution objective was established by combining gravitational theory. In the model calibration, the Ubiquitous Traffic Eyes open-source dataset was screened to obtain the non-forced lane changing segment data, and the parameters of the model were calibrated using the ant colony algorithm. A randomized cross- validation method was used for validation, and the correct rate was used as the evaluation index of model accuracy and generalization ability, which was compared with the traditional model. The results show that when the training-validation ratio is 72%∶28%, 65%∶35%, 57%∶43%, and 50%∶50%, the average correct rate interval is 87.67% to 90.34%, which indicates that the model is robust and feasible. The proposed model shows higher prediction accuracy compared with the traditional model, which can provide a basis for lane selection of the vehicles in the autonomous driving environment.

Key words: intelligent transportation, lane-changing decision, comprehensive competitiveness, autonomous driving; lateral velocity, trajectory data

中图分类号: