基于粒度计算的覆盖算法

The Covering Algorithm Based on Granular Computing

为了更好地解决高维海量数据的分类问题,本文提出一种基于粒度计算的覆盖算法。该算法以粒度计算为理论依据,指出在分析研究某一问题时,可以适当将其属性、论域或者结构粗化,求得某个商空间,在该商空间中抓住事物的本质对其研究,对某些在同一个粗粒度世界无法识别或者彼此特征区别很弱的对象可以换一个粒度世界对其分析,从而全面了解整个问题;以构造性学习算法——覆盖算法为具体实现工具,得到多个商空间中的结果,最终由商空间理论中的函数合成法获得完整结果。实验证明这种基于粒度计算的覆盖算法在解决分类问题时是行之有效的。

In order to solve the classification problems of many dimensions and large amount of samples better, a covering algorithm based on granular computing is put forward in this paper. The algorithm, whose theory is granular computing, points out that the domain, characters or structure of a problem can be coarser properly when it is analyzed, and a corresponding quotient space is gotten; in the quotient space, the essence of problems can be studied. Then the samples, which are unidentified or whose features are unconspicuous, are classified in the different granular worlds easily. And with the function synthesis method in quotient space theory, different results by the covering algorithm of the different granular worlds are combined finally. The experiments show the rationality and feasibility of this algorithm when the classification problems are analyzed.