一种适用于大气污染数据聚类的距离度量  

A Novel Distance Measure for Air Pollution Data Clustering

作　　者：张忠帅 陈梅[1] 杨鹏飞 ZHANG Zhong-shuai;CHEN Mei;YANG Peng-fei(School of Electronics and Information Engineering,Lanzhou Jiaotong University,Lanzhou 730070,China)

机构地区：[1]兰州交通大学电子与信息工程学院,兰州730070

出　　处：《兰州交通大学学报》2022年第2期63-70,共8页Journal of Lanzhou Jiaotong University

基　　金：国家自然科学基金(61762057);甘肃省重点研发计划(21YF5GA053)。

摘　　要：大气污染抽样数据通常包含较多的属性.在使用聚类技术解析污染源、分析污染特征时,如果对数据进行降维处理与标准化处理,会损失相关污染信息,并使污染特征不明显.为解决此问题,提出了一种适用于大气污染数据的相似性度量—大气污染抽样数据距离,它既不需数据预处理,又能体现污染物的浓度变化程度.在仿真实验中,将大气污染抽样数据距离与常用的其他5个距离分别应用于4个传统聚类算法与2个新聚类算法中.仿真结果表明:在同一聚类算法中,使用大气污染抽样数据距离在大多数情况下会比使用其他距离得到更好的聚类结果.Air pollution sampling data usually contain more attributes.When clustering is used to analyze pollution sources and pollution characteristics,if dimension reduction and standardization are carried out on data,relevant pollution information will be lost and pollution characteristics will not be obvious.To solve this problem,a novel similarity measure suitable for air pollution data clustering,PD(air pollution data distance)distance is proposed,which can reflect the variation degree of pollutant concentration but not require data preprocessing.In the simulation experiment,PD distance and the other 5 commonly used distances are applied to the 4 traditional clustering algorithms and the 2 new clustering algorithms respectively.The simulation results show that in the same clustering algorithm,PD distance can get better clustering results than the other distances in most cases.

关 键 词：聚类分析 相似性度量 大气污染抽样数据分析 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]