“两客一危”车辆危险驾驶行为前兆因素识别及预测模型  

Identification and prediction model on precursor factors of dangerous driving behavior for“two types of passenger vehicles and one type of hazardous materials transport vehicle”

作　　者：陈敬松 王艳 周栩佳 周欣 王志甜 杨逢春 李易 王俊骅[5] CHEN Jingsong;WANG Yan;ZHOU Xujia;ZHOU Xin;WANG Zhitian;YANG Fengchun;LI Yi;WANG Junhua(Guangzhou North Second Ring Transportation Technology Co.,Ltd.,Guangzhou Guangdong 510030,China;China Merchants Chongqing Communications Technology Research and Design Institute Co.,Ltd.,Chongqing 400067,China;Logistics Research Center,Shanghai Maritime University,Shanghai 201306,China;School of Transportation,Shanghai Maritime University,Shanghai 201306,China;College of Transportation Engineering,Tongji University,Shanghai 201804,China)

机构地区：[1]广州市北二环交通科技有限公司,广东广州510030 [2]招商局重庆交通科研设计院有限公司,重庆400067 [3]上海海事大学物流研究中心,上海201306 [4]上海海事大学交通运输学院,上海201306 [5]同济大学交通学院,上海201804

出　　处：《中国安全生产科学技术》2025年第3期209-217,共9页Journal of Safety Science and Technology

基　　金：道路交通安全管控技术国家工程研究中心开放课题(2024GCZXKFKT19A);国家自然科学基金项目(52172348,52202419);上海市科学计划委员会科研计划项目(19DZ1202100);北二环高速公路改扩建期间精细化管控策略研究项目(BEH-2023-ZX-082)。

摘　　要：为深入分析“两客一危”车辆的危险驾驶行为,基于上海市“两客一危”的危险驾驶行为数据,梳理包括车道偏离、车距过近、分神驾驶、前向碰撞、疲劳驾驶和其他类,共6类危险行为数据集,并基于多元Logistic模型对危险行为的潜在前兆因素进行识别和分析。研究结果表明:前序时间内的累计危险行为次数、持续驾驶时长和驾驶时间(4∶00~6∶00)对这6种危险行为均具有显著影响,前序的车道偏离、分神驾驶、前向碰撞、疲劳驾驶行为分别对6种危险驾驶行为具有不同的影响效果;对比LSTM,Bi-LSTM与Bi-GRU3种模型的预测效果,可知Bi-LSTM在测试集上的预测效果最好,其准确率比LSTM模型高7百分点,比Bi-GRU模型高4百分点,并且Bi-LSTM和Bi-GRU模型在疲劳驾驶和其他类的预测效果最好。研究结果可为“两客一危”车辆危险驾驶行为的预防、管理以及智能监控系统的开发提供参考,有助于提升道路交通安全水平。In order to analyze the dangerous driving behavior of“two types of passenger vehicles and one type of hazardous materials transport vehicle”,the dangerous driving behaviordata of these special vehicles in Shanghai was analyzed.Six types of dangerous behavior datasets were sorted out,including lane deviation,close vehicle distance,distracted driving,forward collision,fatigue driving and other.The potential precursor factors of dangerous behavior were identified and analyzed based on multivariate Logistic model.The results show that the cumulative number of dangerous behavior,continuous driving duration and driving time period(4∶00~6∶00)in the preceding timehave significant influence on these six types of dangerous behavior,and the preceding lane deviation,distracted driving,forward collision and fatigue driving have different influence on the six types of dangerous driving behavior,respectively.By comparing the prediction effect of LSTM,Bi-LSTM and Bi-GRU models,it can be seen that Bi-LSTM has the best prediction effect on the test set,and its accuracy is 7 percentage points higher than that of LSTM model and 4 percentage points higher than that of Bi-GRU model,and Bi-LSTM and Bi-GRU models have the best prediction effect on fatigue driving and other type.The research results can provide reference and effective prediction model for the prevention and management on dangerous driving behavior of“two types of passenger vehicles and one type of hazardous materials transport vehicle”and the development of intelligent monitoring system,which is helpful to improve the level of road traffic safety.

关 键 词：两客一危 危险驾驶行为 前兆因素 行为预测 

分 类 号：X951[环境科学与工程—安全科学]