模糊信息系统上的粗糙约简  

Rough Reduction in Fuzzy Information Systems

作　　者：管涛[1] 薛亮[1] 冯博琴[1] 

机构地区：[1]西安交通大学电子与信息工程学院,西安710049

出　　处：《西安交通大学学报》2005年第6期574-577,632,共5页Journal of Xi'an Jiaotong University

基　　金：国家高技术研究发展计划资助项目(2003AA1Z2610).

摘　　要：:基于粗糙集理论和模糊等价关系,提出了模糊信息系统(FISs)在不同粒度划分上的知识约简方法和属性重要性度量.这些约简利用了2个水平划分参数(或对象相似度)α、β,其中相对约简与属性重要性度量采用了决策类的水平集正区域公式.利用水平集粗糙成员函数得到分布约简与分配约简方法,它们扩展了Pawlak信息系统(PISs)上的属性约简方法,解决了FISs上的知识获取与特征选择问题.同时,基于不同粒度下的等价类,给出了FISs上可辨识属性矩阵、分布约简和分配约简的辨识公式,克服了经典方法在FISs上的不适用性.示例结果表明,在不同粒度空间上,这些约简方法产生了与全部属性具有最大程度分辨能力和规则置信度的属性子集.Based on rough set theories and fuzzy equivalence relations, a knowledge reduction method and importance degree of attributes under different granular partitions of the object space in fuzzy information systems (FISs) were presented. Two parameters for different level partitions (or similar degree between objects) α, β are used in these reductions, in which the positive region formula of the decision level set are adopted for relative reduction and importance degree of attributes, and the distributed reduction and the assignment reduction are obtained by using rough membership functions of the horizontal set. These reductions extend the attribute deduction methods in Pawlak information systems (PISs) and provide new tools for knowledge discovery and feature selection in FISs. Moreover, by using equivalence classes under different granules, the discernment attribute matrix and discernment formula of the distributed reduction and the assignment reduction were given, which overcome the inapplicability of classical methods in FISs. Demonstration results show that the attribute subsets having maximum degree of discernment and rule confidence with regard to all attributes can be produced by using these methods in different granular spaces.

关 键 词：模糊信息系统 相对约简 分布约简 分配约简 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]