基于数据挖掘处理的影响辽宁高速公路冬季交通事故主要气象要素分析

Analysis of main meteorological factors influencing winter traffic accidents in Liaoning expressway based on data mining processing

利用2014—2016年辽宁省冬季高速公路事故记录,将多要素逐日气象观测数据与事故数据按照日期进行匹配,分析气象条件引发高速事故的空间分布特征。数据挖掘分析首先利用两步聚类方法确定辽宁地区冬季天气类型数量,再通过K-means方法对气象数据进行聚类处理。采用随机森林方法对不同天气类型搭建高速公路交通事故分类模型,并分析模型中气象要素的特征重要性。结果表明:受气象条件影响的高速事故数量辽南地区最多,其次是辽西地区,辽东、辽北地区占比较低。辽宁冬季天气可以分为四种类型,根据气象要素数据结构总结出的天气特征分别为:当日出现降水、前一日出现降水、寒冷干燥、潮湿回暖。有明显降水特征天气类型的事故率超过七成,降温、升温天气类型的事故率在两成左右。随机森林方法对前一日出现降水、寒冷干燥两种天气类型的分类精度更高,模型泛化能力也更好。4种天气类型中气象要素的特征重要性有明显差异,最高地温要素在事故高发天气类型中特征重要性排第1位,在潮湿回暖天气类型排第3位,对于冬季辽宁地区高速公路交通安全影响高于其他要素。

Using the records of winter expressway accidents occurring in Liaoning province and the corresponding multi-factor weather observational data from 2014 to 2016,the spatial distribution characteristics of expressway accidents influenced by meteorological conditions were analyzed.Two-step clustering method was used first in data mining analysis processing to determine the number of winter weather types in Liaoning province,and then the cluster analysis of the meteorological data was performed using K-means method.The random forest method was used to construct an expressway traffic accident classification model for different weather types,and finally,the characteristic importance of meteorological factors in the model was analyzed.The results show that the number of expressway accidents influenced by meteorological conditions is the highest in the southern Liaoning province,followed by the western area,and numbers in the eastern and northern area are relatively low.The winter weather in Liaoning province can be divided into four types.According to the data structure of meteorological factors,the weather characteristics are as follows:precipitation occurring on the day,on the previous day,cold and dry,and warm and humid.The accident rate of weather types with obvious precipitation characteristics exceeds 70%,and the one with cooling and warming weather types is about 20%.The random forest method has high classification accuracy for the two weather types,i.e.,precipitation occurring on the day and cold and dry types.The generalization ability of the model is better.The characteristic importance parameters of meteorological factors for the four weather types are significantly different.The surface temperature factor has a greater significance,occupying the first place in the high-incidence weather types,and the third place in the humid warm weather type.