摘要
为确保复杂线形组合路段车辆行驶安全,针对当前公路线形信息量化表达不足,以及路段几何线形与驾驶人视觉线形对小客车行驶速度耦合影响考虑不充分的问题,提出基于数据驱动的高速公路小客车行驶速度分布预测模型。基于八自由度驾驶模拟器和高精度眼动仪开展试验,采集38名驾驶人在106个山区高速公路典型路段的小客车行驶速度和视觉感知数据。从“路段-视野”双重视角出发提取路段几何线形特征,采用变分自编码器进行驾驶人视觉线形重构,提出“路段-视野”图结构以融合和量化不同视角的线形信息,基于时空图注意力神经网络构建小客车行驶速度分布预测模型。结果表明:模型预测性能优异,行驶速度均值和标准差的预测准确度分别达到96.3%和98.1%,均方根误差分别为4.6、1.2km·h^(-1),平均绝对误差分别为3.8、0.8km·h^(-1)。此外,消融试验和对比试验结果进一步验证了模型在不同典型路段的有效性和实用性。通过自注意力机制发现,路段几何线形对小客车行驶速度分布在时间上具有持续的引导作用,而驾驶人视觉线形在空间上对其进行动态调节和反馈。研究结果可为智能化的公路安全性评价和几何设计方法提供理论基础,并有助于精细化的智能交通系统速度管理。
To ensure driving safety on highway segments with complex alignment combinations,this study proposed a data-driven prediction model for passenger car speed distribution.The model addressed the issues of insufficient quantitative representation of highway alignment information and the limited consideration of the coupling effect between highway geometric alignment and drivers'visual perception on running speed.The study conducted experiments using an eight-degree-of-freedom driving simulator and a high-precision eye-tracking device,collecting driving speed and visual perception data from 38 drivers across 106 typical mountainous highway segments.From the dual perspective of highway segments and drivers'visual field,highway alignment features were extracted,and driver visual alignment was reconstructed using a variational autoencoder model.The study introduced a“highway segment-visual field”graph to fuse and quantify alignment information,and a passenger car speed distribution prediction model was constructed based on spatio-temporal graph attention neural networks.The results indicate that the proposed model exhibits superior performance on the testing dataset,with a prediction accuracy of 96.3%and 98.1%for the mean and standard deviation of running speed,respectively.The root mean square errors are 4.6,1.2 km·h^(-1),while the mean absolute errors are 3.8,0.8 km·h^(-1).Additionally,ablation and comparative experiments further validate the model's effectiveness and applicability across different typical highway segments.The self-attention mechanism reveals that highway geometric alignment provides a fundamental temporal guide for passenger car speed,while drivers'visual alignment offers dynamic spatial adjustment and feedback.The findings could establish a theoretical foundation for intelligent highway safety evaluation and geometric design methods,while also contributing to the advancement of refined speed management in intelligent transportation systems.
作者
高健强
陈雨人
余博
任蔚溪
陈修和
GAO Jian-qiang;CHEN Yu-ren;YU Bo;REN Wei-xi;CHEN Xiu-he(College of Transportation,Tongji University,Shanghai 201804,China;Key Laboratory of Road and Traffic Engineering,Ministry of Education,Tongji University,Shanghai 201804,China;Anhui Transport Consulting&Design Institute Co.Ltd.,Hefei 230088,Anhui,China)
出处
《中国公路学报》
北大核心
2025年第9期377-390,共14页
China Journal of Highway and Transport
基金
国家重点研发计划项目(2023YFE0202400)
上海市自然科学基金项目(22ZR1466000)。
关键词
交通工程
行驶速度分布
时空图注意力神经网络
路段几何线形
驾驶人视觉线形
驾驶模拟试验
traffic engineering
running speed distribution
spatio-temporal graph attention neural networks
highway geometric alignment
driver visual alignment
driving simulator experiment
