Abstract: Knowledge reduction is one of the most important contributions of rough set theory （RST）. At present， many kinds of knowledge reduction algorithm based on RST are proposed. Mostly research focuses on attribute reduction. Based on the definition of bit granular matrix （BGrM）， a matrix-based knowledge discovery algorithm is proposed， and it can realize reduction both for attribute and attribute value. Furthermore， the line by line attribute value reduction is improved to category reduction. Examples are given to illustrate the proposed approach.

Key words: granular computing （GrC）, rough set theory （RST）, bit granular matrix （BGrM）, knowledge reduction

摘要： 知识约简是粗糙集理论精髓之一，目前有多种基于粗糙集理论的知识约简算法，相关研究基本集中在属性约简，对属性值的约简研究相对较少。在定义粒矩阵和矩阵运算的基础上，提出了基于粒矩阵的知识约简方法。它既能进行属性约简也能进行属性值约简，旨在把对属性值的逐行约简转化为对范畴的直接简化。算例表明了基于粒矩阵的知识约简方法的基本思想。

关键词: 粒计算, 粗糙集理论, 二进制粒矩阵, 知识约简