一种基于局部密度的网格排序聚类算法

摘要

In the process of clustering,numerous complex data appear with noise.This paper proposed a clustering method which utilized a grid scheduling strategy based on local density (GSS-LD).Firstly,GSS-LD used the local properties of clus-tering sorting grid cell,and transformed the grid clustering problem into grid scheduling problem.Secondly,it used the con-cept of relative local density gradient to overcome the limitations of global parameters in conventional clustering algorithm. Thus GSS-LD held a strong ability to deal with the clustering of multi-density data set with noise.The comparing results on 3 data sets with other 2 methods show that GSS-LD is feasible and to handle with the clustering for multi-density data.Mean-while,time complexity of GSS-LD has a linear relation with the scale of data set.%针对复杂及带噪声的数据集的聚类问题，提出了一种基于局部密度的网格排序策略（GSS-LD），并以其作为网格聚类的组织模式。GSS-LD利用聚类的局部性质进行网格单元排序，将基于网格的聚类问题转换为网格的排序问题，运用相对局部密度变化率的概念，克服了传统网格聚类算法中全局性参数的局限性，使其可以适应多密度数据集的聚类。通过三组具有不同拓扑结构的数据集测试GSS-LD的聚类性能并与其他两种方法进行比较，结果表明GSS-LD可以对复杂数据集进行有效聚类，其时间复杂度分别与数据规模及网格结构具有线性关系，同时具有较强的噪声处理能力。