| 基于道路线形的智能汽车事故多发路段预判模型 |
| |
| 引用本文： | 宁航,赵祥模,南春丽,杨澜,李婧.基于道路线形的智能汽车事故多发路段预判模型[J].中国公路学报,2021,34(3):183-192. |
| |
| 作者姓名： | 宁航 赵祥模 南春丽 杨澜 李婧 |
| |
| 作者单位： | 长安大学 信息工程学院, 陕西 西安 710064 |
| |
| 基金项目： | 国家重点研发计划项目(2018YFB0105104);国家自然科学基金联合基金项目(U1864204);国家自然科学基金项目(61703053);陕西省博士后科研项目(2018BSHYDZZ64)。 |
| |
| 摘 要： | 分析了道路线形对智能汽车行驶安全性的影响，分别使用数据驱动的机器学习方法和模型驱动的经典数学建模方法，建立了以道路线形技术指标为输入的神经网络模型和多元数学模型，预测事故多发路段；计算了各个道路线形技术指标与事故率之间的偏相关系数，从中挑选出与事故率相关程度较大的道路线形特征，使用T检验和F检验验证了道路线形特征组合和单个特征对事故率的影响。结果表明：基于机器学习的神经网络模型和基于数值逼近理论的多元数学模型预测正确率基本相近，大约为90%；2种模型对道路安全影响较大的道路线形相关不利因素组合相同，均为平曲线转角、横向力系数和纵坡坡度；各种不利因素组合中，平曲线转角、横向力系数和纵坡坡度出现的频率分别为100.0%、91.7%和83.3%，远远大于其他因素；事故多发路段道路线形因素不仅与平曲线转角、横向力系数和纵坡坡度有关，而且与其线形组合有密切关系，组合不当亦会导致事故增加；2种模型可相互验证，考虑计算速度及参数的可解释性，实际中应优先选择多元数学模型进行事故预判。
|
| 关 键 词： | 交通工程 事故多发路段 数据驱动 行驶安全 智能汽车 道路线形 预判模型 模型驱动 |
| 收稿时间： | 2020-11-18 |
Accident-prone Section Prediction Models for Intelligent Vehicles Based on Road Alignment |
| |
| NING Hang,ZHAO Xiang-mo,NAN Chun-li,YANG Lan,LI Jing.Accident-prone Section Prediction Models for Intelligent Vehicles Based on Road Alignment[J].China Journal of Highway and Transport,2021,34(3):183-192. |
| |
| Authors: | NING Hang ZHAO Xiang-mo NAN Chun-li YANG Lan LI Jing |
| |
| Affiliation: | School of Information Engineering, Chang'an University, Xi'an 710064, Shaanxi, China |
| |
| Abstract: | The impact of road alignment on the driving safety of intelligent vehicles was analyzed in this study.Using data-driven machine learning and model-driven classic mathematical modeling,a neural network model with road alignment parameters and a multivariate mathematical model were built as the input respectively to predict accident-prone sections.The partial correlation coefficients of all the road alignment parameters and accident rates were calculated,from which the characteristics of road alignments with higher levels of relevance to accident rates were selected.The effects of combinations of such characteristics and of individual characteristics on the accident rates were then verified through a T-test and F-test.According to the results,the machine learning-based neural network model and numerical approximation-based multivariate mathematical model have similar prediction accuracies,about 90%.Both models identify the same combination of road-alignment-related adverse factors that significantly impact road safety,the factors are horizontal curve corner,lateral force coefficient,and longitudinal gradient.Among the combinations of adverse factors,the frequencies of the horizontal curve corner,lateral force coefficient,and longitudinal gradient are 100.0%,91.7%,and 83.3%respectively,which are much higher than those of other factors.The road alignment factor in accident-prone sections is related to not only the three aforementioned factors but also the road alignment combinations,and inappropriate combinations will also increase the number of accidents.The two models can verify each other.Considering the computation speed and interpretability of the parameters,the multivariate mathematical model is preferred in practice for accident prediction. |
| |
| Keywords: | traffic engineering accident prone section data-driven driving safety intelligent vehicle road alignment prediction model model-driven |
| 本文献已被 CNKI 维普 等数据库收录！ |
| 点击此处可从《中国公路学报》浏览原始摘要信息 |
|
点击此处可从《中国公路学报》下载全文 |
|
