交通运输系统工程与信息

交通运输系统工程与信息 ›› 2025, Vol. 25 ›› Issue (3): 96-106.DOI: 10.16097/j.cnki.1009-6744.2025.03.009

• 智能交通系统与信息技术 • 上一篇    下一篇

山区双车道公路借道超车轨迹预测模型

覃文文1a,1b，彭栋梁1a，戢晓峰1a,1b，徐迎豪1a，李冰1a，李武*1a,1b，曾浩2   

  1. 1. 昆明理工大学，a.交通工程学院，b.云南省现代物流工程研究中心，昆明650504；2. 昆明元朔建设发展有限公司，昆明650228
  • 收稿日期:2024-08-12 修回日期:2024-10-07 接受日期:2024-10-21 出版日期:2025-06-25 发布日期:2025-06-20
  • 作者简介:覃文文(1986—)，男，广西柳州人，讲师。
  • 基金资助:
    国家自然科学基金(72461015)；云南省基础研究计划面上项目(202401AT070309)；云南省研究生导师团队建设项目(2024)。

Lane-occupying Overtaking Trajectory Prediction Model for Two-lane Mountainous Highways

QIN Wenwen1a,1b, PENG Dongliang1a, JI Xiaofeng1a,1b, XU Yinghao1a, LI Bing1a, LI Wu*1a,1b, ZENG Hao2   

  1. 1a. Faculty of Transportation Engineering, 1b. Yunnan Modern Logistics Engineering Research Center, Kunming University of Science and Technology, Kunming 650504, China; 2. Kunming Yuanshuo Construction Development Co Ltd, Kunming 650228, China
  • Received:2024-08-12 Revised:2024-10-07 Accepted:2024-10-21 Online:2025-06-25 Published:2025-06-20
  • Supported by:
    National Natural Science Foundation of China(72461015)；Yunnan Fundamental Research Project Projects(202401AT070309)；Yunnan Postgraduate Supervisor Team Construction Project (2024)。

PDF

摘要： 为提高山区双车道公路的车辆轨迹预测精度，本文设计一种考虑借道超车影响的车辆轨迹预测模型。首先，基于无人机视频轨迹数据，根据航向角将借道超车过程划分为跟驰、借道、超车和返回这4种状态；其次，构建包含借道超车状态、车辆运动特征、空间位置属性和交通状态的多元特征数据集，通过梯度提升决策树算法拟合借道超车状态与车辆运动特征、空间位置和交通状态之间的复杂关系，采用SHAP(SHapley Additive exPlanations)方法识别影响借道超车状态变化的关键因素；然后，将借道超车状态、影响借道超车状态的关键因素和历史轨迹数据集，以滑动时间窗口形式输入至Informer模型，预测山区双车道公路的借道超车轨迹；最后，与未考虑借道超车影响的传统超车轨迹预测模型进行对比，验证本文模型的有效性。结果表明：车头时距、主体车辆横向速度和横向偏移是影响借道超车状态变化的关键因素；所构建的模型在山区双车道借道超车场景下，具有良好的适用性和预测精度；与未考虑借道超车影响的轨迹预测模型相比，本文模型的均方误差和平均绝对误差均值分别降低53.05%和38.11%，决定系数均值提升23.58%。

关键词: 交通工程, 超车轨迹预测, Informer时间序列模型, 借道超车, 山区双车道

Abstract: To enhance the accuracy of vehicle trajectory prediction on two-lane mountainous highways, this paper proposes a trajectory prediction model that considers the impact of lane-occupying overtaking. First, based on the Unmanned Aerial Vehicle video trajectory data, the lane-occupying overtaking process is divided into four states according to the heading angle: following, lane-occupying, overtaking, and returning. Second, a multivariate feature dataset is constructed, containing lane-occupying states, vehicle motion characteristics, spatial position attributes, and traffic conditions. The Gradient Boosting Decision Tree (GBDT) algorithm is then used to fit the complex relationships between lane-occupying states and vehicle motion characteristics, spatial positions, and traffic conditions. The SHAP (SHapley Additive exPlanations) method is used to identify key factors affecting lane occupying state changes. Then, the lane-occupying states, key factors influencing these states, and historical trajectory datasets are input into the Informer model in the form of a sliding time window to predict lane-occupying overtaking trajectories on two-lane mountainous highways. The effectiveness of this model is verified by comparing it with traditional overtaking trajectory prediction models that do not consider lane-occupying impacts. The results indicate that time headway, lateral speed of the main vehicle, and lateral offset are key factors affecting lane-occupying state changes. The proposed model demonstrates good applicability and prediction accuracy in the context of lane-occupying overtaking on two-lane mountainous highways. Compared with the trajectory prediction model without considering the effect of borrowed lane overtaking, the mean square error and mean absolute error of the model in this paper are reduced by 53.05% and 38.11%, respectively, and the mean value of the coefficient of determination is improved by 23.58%.

Key words: traffic engineering, overtaking trajectory prediction, Informer time series forecasting, lane-occupying behavior; mountainous dual carriageway

中图分类号: