%0 Journal Article
%T 邻域密度网格聚类算法及应用<br>Neighborhood density grid clustering and its applications
%A 索明亮
%A 周鼎
%A 安若铭
%A 李顺利
%J 清华大学学报（自然科学版）
%D 2018
%R 10.16511/j.cnki.qhdxxb.2018.22.025
%X 聚类作为数据分析的工具之一，已在模式识别、文献计量及故障诊断等领域中发挥了重要作用。该文基于邻域关系、局部密度和空间网格划分提出了一种聚类方法。该方法主要利用空间网格降低计算复杂度，利用邻域关系在网格空间中以密度为依据搜索聚类元素，并根据最大相对距离和最大相对密度原则自动寻找聚类中心。基于人工数据的实验结果表明，所提邻域密度网格聚类方法可有效处理任意形状数据并自主完成聚类。基于区域识别的对比实验表明，所提方法更适用于处理奇异形状且分布复杂的数据。<br>Abstract：The clustering data analysis tool plays a significant role in various fields such as pattern recognition, bibliometrics and fault diagnosis. This paper describes a clustering approach based on neighborhood relationships, local densities and spatial grid partitions. The time complexity of this algorithm is reduced using a spatial grid with the clustering elements searched using neighborhood density relationships in the grid space. Cluster centers are then selected automatically using the maximum relative distance and the maximum relative local density. Tests on artificial data indicate that neighborhood density grid clustering can automatically cluster data and effectively process data with arbitrary shapes. Comparisons using regional recognition datasets demonstrate that this method is more suitable for clustering complex data with unusual shapes.
%K 聚类
%K 网格划分
%K 邻域
%K 密度
%K 任意形状
%K 区域识别
%K &lt
%K br&gt
%K clustering
%K grid partition
%K neighborhood
%K density
%K arbitrary shape
%K region recognition
%U http://jst.tsinghuajournals.com/CN/Y2018/V58/I8/732