基于改进聚类算法的道路交通事故多发路段鉴别方法研究

Accident-prone Section Identification Approach Based on Improved Clustering Algorithm

针对现有道路事故多发段鉴别方法在阈值选择中存在的缺点和局限,提出了一种改进的DBSCAN聚类算法用于鉴别事故多发段.聚类算法中最小密度点的选取结合了累计频率曲线方法,能实现阈值的自适应选取.算法应用于安徽省某高速公路,结果表明采用的聚类算法能实现任意长度的聚类,不会遗漏事故多发段或扩大事故多发段范围,且识别的事故多发路段更为集中,是一种鉴别事故多发段空间分布特征的有效方法.

Based on the analysis on limitation of present common road accident-prone location identification method,an improved DBSCAN clustering algorithm was proposed to identify the accident-prone location.In the algorithm,the minimal threshold value of clustering is adaptively changed in combination with cumulative frequency curve method,where the optimal value can be found.The application in a highway of Anhui province shows that,the improved DBSCAN clustering algorithm not only can identify the clustering of highway accident-prone area with any section length,but also be able to make the section length more concentrate.In addition,it will not miss any accident-prone section or expand the accident-prone area.It is proved to be an effective method to studying spatial distribution characteristics of accident-prone location.