改进K-means的空间聚类算法

Improved K-means clustering algorithm on space

提出了基于K-means的四叉树与R-link树的混合结构树,提高了R-link树的查询性能,在K-means中采用均值—标准差确定初始聚类中心,提高了收敛速度,通过距离准则函数来优化K值,避免K值的盲目选取。与R-link相比空间开销代价有时略大,但换取了更高的性能,且数据量越多,此种结构的整体性能越好,适合于海量数据。

This paper presented a quick speed spatial indexing structure which was based on R-link tree, And it used K- means algorithm in the structure. In K-means algorithm, adopted value-standard deviation to ascertain the initial clustering centres to improve convergence speed and ascertain ultimate K value by distance criterion .function to make K value most suitable. The structure sometimes consumes more storage than R-link but gains better performance. Furthermore, data quantity more, this kind of structure overall performance is better.