和声搜索的半监督聚类研究与应用

摘要

As there are some problems of the existing clustering algorithms, how to fastly find the optimal cluster centers by harmony search is studied Some improvements is also carried out for the harmony search algorithm. For getting the best number of cluster center, a semi-supervised method is used to test clustering quality under variable number of centers iteratively. Considering the different influences of each dimension attribute on the clustering effect, each dimension is weighted to select feature. All of these methods is to achieve a better clustering quality. Experimental results show that the proposed clustering algorithm outperforms other similar algorithms. Finally, the proposed algorithm is applied to parallel computing performance analysis to distinguish and identify the performance category of various processor in parallel computer.%由于现有的聚类算法还存在一些问题,研究了如何用和声搜索算法快速寻找最优的聚类中心,对于和声搜索算法也进行了一些改进.为了获得最佳的类中心数,采用了半监督方式循环测试各种中心数情况下的聚类质量.考虑到各维特征属性对聚类效果影响不同,采用了维度加权的方法进行特征选择.所有这些措施都是为了达到一个更好的聚类效果.实验结果表明,该聚类算法性能优于其它同类算法.算法被应用于并行计算性能分析中,用于区分和识别并行机的各个处理器运行性能类别.