摘要
为了精准有效地进行交通事故预防预警,基于车辆OBD驾驶行为数据及信息熵理论,提出了城市道路交通安全风险预估方法。首先,分析异常驾驶行为高发位置与道路交通事故发生位置的关联性;其次,构建以道路交通安全熵为一级指标,急加速率、急减速率、急转弯率、超速率、高速空挡滑行率为二级指标的道路交通安全风险预估指标体系,提出了基于改进熵权法的道路交通安全熵计算方法;然后,基于密度聚类、K-means聚类提出了道路交通安全风险等级数确定方法,并基于K-means聚类建立了风险等级阈值计算方法。研究结果表明:异常驾驶行为高发位置与交通事故发生位置具有一致性;通过对log对数底数选择优化、二级指标零值处理、指标权重分段计算3个步骤改进的熵权法,可弥补log对数函数无法计算零值指标熵值的缺陷,避免指标权重为负、指标熵值与权重反映信息不一致的现象;两步聚类避免了孤立数据点对安全风险等级划分的影响。以重庆市4条城市道路(总长约38km)进行实例验证后得出,道路交通安全熵与交通事故表征的道路交通安全状态趋势一致;道路交通安全风险等级可划分为高、低风险2级,道路交通安全熵优化阈值为0.042,最后,风险等级划分准确率为87.88%。研究成果可为道路交通安全风险点辨识、交通事故预防预警提供有效的技术支持。
For the precise and effective prevention of traffic accidents by providing preemptive warnings,an urban road traffic safety risk estimation method is proposed based on driving behaviors obtained from vehicle OBD data and by using information entropy theory.First,the relationship between locations where abnormal driving behaviors occur frequently and locations where traffic accidents occur was analyzed.Second,a road traffic safety risk evaluation index system was developed;the system included a first class index,such as road traffic safety entropy,and second class indexes such as the rapid acceleration frequency,rapid deceleration frequency,swerve frequency,speeding frequency,and high-speed idling frequency.Thus,a road traffic safety entropy calculation model was developed based on improving the entropy weight method.Then,the number of road traffic safety levels was determined by combining density clustering and K-means clustering.Furthermore,a method for calculating safety level thresholds was established through K-means clustering.The test results demonstrate the following:①locations where high-frequency abnormal driving behaviors occur are consistent with the locations of occurrence of traffic accidents;②modifications to the entropy weight method,such as optimization of the selection of the base of the logarithmic function,the processing of second class indexes that are equal to zero,and the piecewise calculation of index weights,can compensate for the limitation of the original logarithm with regard to entropy calculation with indexes equal to zero;in addition,it can avoid negative index weights and information inconformity between entropy values and index weights;③two-step clustering may evade the influences of isolated data points on safety level classification.A verification study was carried out on four urban roads in Chongqing that had a total length of approximately 38 km;the results demonstrate that the road traffic safety entropy is in accordance with the trend of the road safety status reflected by accidents,that road traffic safety risk can be classified into high and low levels,and that the optimal road traffic safety entropy threshold is 0.042.Finally,the model achieves a risk classification accuracy of 87.88%.This research can provide effective technical support for the identification of risky locations with regard road traffic safety,as well as for preventing traffic accidents by providing preemptive warnings.
作者
蔡晓禹
雷财林
彭博
唐小勇
高志刚
CAI Xiao-yu;LEI Cai-lin;PENG Bo;TANG Xiao-yong;GAO Zhi-gang(Chongqing Key Lab of Traffic System&Safety in Mountain Cities,Chongqing 400074,China;College of Traffic and Transportation,Chongqing Jiaotong University,Chongqing 400074,China;Chongqing Urban Transportation Big Data Engineering Technology Research Center,Chongqing 400020,China)
出处
《中国公路学报》
EI
CAS
CSCD
北大核心
2020年第6期190-201,共12页
China Journal of Highway and Transport
基金
重庆市技术创新与应用示范专项社会民生类重点研发项目(cstc2018jscx-mszdX0085)
国家自然基金青年科学基金项目(61703064)
重庆市城市交通大数据工程技术研究中心科研项目(SW-2018-Z016)。
关键词
交通工程
风险预估
信息熵
驾驶行为
熵权法
两步聚类
交通安全
traffic engineering
risk estimate
information entropy
driving behavior
entropy weight method
two-step clustering
traffic safety
