自适应邻域密度聚类及事故黑点识别应用  

Density Clustering with Adaptive Neighbors and Its Application to Accident Black Spots Recognition

作　　者：刘韡 黄俊龙 鲁娜 刁麓弘[2] LIU Wei;HUANG Junlong;LU Na;DIAO Luhong(Aerospace Science and Industry Network Information Development Co.,Ltd.,Beijing 100080,China;Faculty of Science,Beijing University of Technology,Beijing 100124,China)

机构地区：[1]航天科工网络信息发展有限公司,北京100080 [2]北京工业大学理学部,北京100124

出　　处：《黑龙江交通科技》2024年第6期138-143,150,共7页Communications Science and Technology Heilongjiang

基　　金：北京市教委科技计划一般项目(KM201910005013)。

摘　　要：聚类作为识别交通事故黑点的主要方法之一,其主要问题是交通事故多发区事先无法确定,即无法提前知道聚类簇数。利用样本点之间的连接概率定义了数据点的局部密度,根据局部密度大小来确定聚类中心和簇数,再对数据点进行聚类。结果表明:一是算法对参数不敏感,具有较好的通用性;二是算法能自动确定聚类簇数;三是算法聚类过程只依赖局部密度与邻接点,能够识别噪声点,提升结果的准确性。运用算法在一些真实数据集上进行试验,将聚类结果与其他算法结果利用评价指标ARI(Adjusted Rand Index)和NMI(Normalized Mutual Information)进行比较。最后利用算法对美国6个州的交通事故进行聚类,结果表明算法对交通事故有较好的适应性,能将城市及周边道路上事故密集区域准确识别出来。Clustering is one of the main methods for recognizing large-scale traffic accident black spots.However,its main problem is that the number of traffic accident-prone areas cannot be determined in advance,which means the number of clusters cannot be known.The paper adopts adjacent probability to define the density of data points.Based on it,the clustering centers are determined.Then the data points are clustering into groups according to their relationship with the cluster centers.The results show that:firstly,the algorithm is insensitive to the parameters,which means it is practicable in general;secondly,the algorithm can automatically determine the number of clusters;thirdly,the algorithm's clustering process only relies on local density and neighbor points,which can identify noisy points and improve the accuracy of the results.The proposed algorithm are tested on some real datasets,and the clustering results are compared with the results obtained by other algorithms using evaluation indexes ARI(Adjusted Rand Index)and NMI(Normalized Mutual Information).The algorithm is then used to cluster traffic accidents on the data of six US states,the experimental results indicate that the algorithm has a good adaptability to traffic accidents and can find the traffic accident-prone areas well.

关 键 词：交通事故黑点 聚类算法 聚类簇数 自适应邻域聚类 局部密度 

分 类 号：U492[交通运输工程—交通运输规划与管理]