基于监控数据的高速公路实时事故风险模型

近年来,高速公路事故发生率高居不下.同时,对于高速公路而言,其交通流检测器安装又较为普遍.因此研究如何深入挖掘交通流检测数据以实现对高速公路事故风险实时预测很有必要.基于美国加州2012年发生事故最多的4条高速公路I5,I10,I405和I15的全年事故数据和交通流数据,以病例对照基本思路选取事故组和对照组数据,选定交通流数据研究范围,并选用ADASYN算法处理不平衡数据集问题.基于随机森林模型,利用事故发生前10~40 min内的事故地上游4个检测器、下游2个检测器的3种基本交通流数据构建高速公路实时事故风险模型,事故预测准确率可达到88.02%.选取重要性前十的变量作为事故重要诱导因素,对事故重要诱导因素进行调值,将调值后的测试集放入之前构建的随机森林模型进行分类预测,结果显示减少了41.82%的事故,故可认为利用事故重要诱导因素可进行事故先兆预警工作,从而减少事故的发生. 

A Real-time Accident Risk Model on Freeways Based on Monitoring Data

In recent years,the incidence of highway accidents on freeways remains high.In the meantime,loop detectors are commonly equipped on freeways.Thus,it is necessary to dig the data of loop detectors in order to predict realtime risk of traffic accidents on freeways.Based on data of actual accidents and collected from detectors on four freeways called I5,I10,I405 and I15 in California,where the most accident numbers occurred in the year 2012,extracting data group of accidents and non-accidents based on an idea of case-control study.Study coverage of detector data is selected.Meanwhile,ADASYN algorithm is used to solve the problem of unbalanced data sets.Based on random forest,three basic traffic flow data within 10-40 min before accidents collected from four upstream detectors and two downstream detectors is used to compute locations of accidents.A real-time accident risk model on freeways is developed with the accuracy rate of accident prediction is 88.02 %.The top ten important variables are selected as important inducements of accidents.Then,the values of the important inducements are adjusted.The modified test set is applied to the random forest model for classification forecasting afterwards.The result shows that the numbers of accidents are reduced by 41.82%.Therefore,it can be found that the important inducements of accidents can be applied to the early warning of traffic accidents,thus reducing the incidence of them.