交通运输系统工程与信息 ›› 2025, Vol. 25 ›› Issue (1): 221-230.DOI: 10.16097/j.cnki.1009-6744.2025.01.021

• 系统工程理论与方法 • 上一篇    下一篇

高速公路跟车情景下认知分心影响机制与识别方法

彭金栓,张淋俊,周磊,袁浩,任超宇,徐磊   

  1. 重庆交通大学,交通运输学院,重庆400074
  • 收稿日期:2024-11-30 修回日期:2024-12-17 接受日期:2024-12-23 出版日期:2025-02-25 发布日期:2025-02-24
  • 作者简介:彭金栓(1982—),男,安徽太和人,教授,博士。
  • 基金资助:
    教育部人文社会科学研究规划基金(24YJAZH110);重庆市高校创新研究群体项目(CXQT21022);重庆市自然科学基金(CSTB2022NSCQ-MSX0549)。

Influence Mechanisms and Identification of Cognitive Distraction of Car-following on Expressways

PENG Jinshuan, ZHANG Lingjun, ZHOU Lei,YUAN Hao, REN Chaoyu, XU Lei*   

  1. College of Traffic & Transportation, Chongqing Jiaotong University, Chongqing 400074, China
  • Received:2024-11-30 Revised:2024-12-17 Accepted:2024-12-23 Online:2025-02-25 Published:2025-02-24
  • Supported by:
    Humanities and Social Sciences Research Planning Fund Project of the Ministry of Education (24YJAZH110);University Creative Research Group Project of Chongqing(CXQT21022);Natural Science Foundation of Chongqing (CSTB2022NSCQ-MSX0549)。

摘要: 为精细化研究认知分心对高速公路场景下驾驶人跟车行为的影响机理,设计面向不同分心次任务的模拟驾驶试验。动态采集车辆运动学特性,驾驶人操作和眼动特征参数,解析次任务状态与速度区间对跟车绩效的影响机制,构造面向不同速度区间跟车行为的认知分心状态表征参数集合。引入支持向量机、随机森林和极端梯度提升树等方法,实时识别驾驶人的认知分心状态。研究表明:沉浸式计算相较于交谈次任务会给驾驶人带来更大的认知负荷;认知分心导致驾驶人对方向盘和油门踏板的控制能力减弱,注视点更加集中于道路前方,视觉转移受到抑制;不同速度区间下,认知分心表征参数集合存在差异;极端梯度提升树模型性能优于支持向量机和随机森林;标定不同速度区间下的最佳滑动时窗宽度与滑动步长,极端梯度提升树模型在总体区间及速度区间Ⅰ([60,80)km·h-1)、Ⅱ([80,100)km·h-1)、Ⅲ([100, 120] km·h-1)下识别准确率分别达到85.98%、87.98%、88.45%、92.21%;截至风险阈值时刻,认知分心样本识别率最高可达90.0%。研究结果可为高速公路认知分心识别及预警优化设计等提供重要参考。

关键词: 交通工程, 认知分心识别, 极端梯度提升树, 高速公路跟车, 驾驶模拟, 识别时序性

Abstract: To investigate the impact of cognitive distraction on drivers' car-following behavior on expressways, this study conducted driving simulation experiment with various distraction tasks. The study dynamically collected vehicle kinematics characteristics, driver manipulation, and eye movement parameters, and analyzed the influence mechanism of the secondary task state and speed interval on car-following performance. A set of cognitive distraction state representation parameters was developed for car-following behavior in different speed intervals. Methods such as the Support Vector Machine (SVM), Random Forest (RF), and Extreme Gradient Boosting (XGBoost) were introduced to identify drivers' cognitive distraction states in real-time. The findings indicated that immersive computing imposed a higher cognitive load on drivers compared to conversational secondary tasks. Cognitive distraction reduced drivers' control over the steering wheel and throttle pedal, more focused gaze on the road ahead, and suppressed visual transfer. The cognitive distraction representation parameters varied across different speed intervals. The XGBoost model outperformed both the SVM and RF. By calibrating the optimal sliding window width and step size under different speed intervals, the XGBoost model achieved recognition accuracies of respectively 85.98%, 87.98%, 88.45%, and 92.21% for the overall interval and the speed intervals of I ([60, 80) km·h-1), II [80, 100) km·h-1), and III [100, 120] km·h-1). Up to the risk threshold moment, the recognition rate for cognitive distraction samples reached a maximum of 90%. The findings provide references for recognizing cognitive distraction and optimizing early warning systems on expressways.

Key words: traffic engineering, cognitive distraction recognition, extreme gradient boosting, expressway car following, driving simulation, identification of time-sequence characteristics

中图分类号: