中文摘要：

属性约简是粗糙集理论一个重要的研究问题．在粗糙集理论上，利用粒计算的思想构建了粒矩阵，提出并定义了粒矩阵相与运算，建立了基于粒矩阵的知识粒化方法，并且给出了粒矩阵属性约简的启发式算法．采用粒矩阵进行属性约简选择最小属性集，跳出了传统属性约简的先求解属性核，再求解最优属性集的方法．理论分析表明了新的算法是可靠有效的，给粒计算属性约简提供一个新的思路，为进一步研究粒计算提供可行的方法．

英文摘要：

Attribute reduction is one of important issues in rough set theory. Based on rough set theory, this paper establishes the granular matrix with the idea of granular computing, defines the AND operation of granular matrix, presents the knowledge granulation method based on granular matrix and proposes an attribute reduction algorithm. The attribute reduction, using granular matrix to select the minimal attribute set, is different from the traditional attribute reduction which acquires the attribute kernel at first and then selects the best attribute set. Theoretical analysis shows that the new algorithm is reliable and valid. The algorithm could provide a new paradigm for the attribute reduction of granular computing and a feasible method for further research on granular computing.