一种改进的悲观多粒度粗糙集粒度约简算法  

An improved granular reduction algorithm for pessimistic multi-granularity rough sets

作　　者：谢立 叶军[1,2] 赖鹏飞 卢岚 周浩岩 李兆彬 XIE Li;YE Jun;LAI Pengfei;LU Lan;ZHOU Haoyan;LI Zhaobin(College of Information Engineering,Nanchang Institute of Engineering,Nanchang 330099,Jiangxi,China;Jiangxi Province Key Laboratory of Water Information Cooperative Sensing and Intelligent Processing(Nanchang Institute of Engineering),Nanchang 330099,Jiangxi,China)

机构地区：[1]南昌工程学院信息工程学院,江西南昌330099 [2]江西省水信息协同感知与智能处理重点实验室(南昌工程学院),江西南昌330099

出　　处：《山东大学学报(工学版)》2024年第6期38-48,56,共12页Journal of Shandong University(Engineering Science)

基　　金：江西省教育厅科技资助项目(GJJ211920);国家自然科学基金资助项目(61562061)。

摘　　要：针对以粒度内部重要度和粒度外部重要度不能有效度量非核粒度的重要度,无法获得有效启发信息使约简过早收敛的问题,提出以正域变化度量核粒度的重要度、以边界集变化度量非核粒度的重要度。新的度量方法不仅能度量核粒度的重要度,而且能度量非核粒度的重要度。以新的粒度重要度为依据,提出一种改进的悲观多粒度约简算法,与样本选择的启发式属性约简算法、信息熵的模糊ε-近似约简算法、粒度加速求解约简算法和邻域区分指数的特征选择算法相比,新算法可以减少迭代次数,能更有效地找到粒度约简子集。通过加州大学欧文分校(University of California Irvine, UCI)数据集进行试验,验证了算法的有效性和实用性。Aiming at the issue that non-kernel granularity significance couldnot be effectively measured by internal and external granularity significance,leading to premature convergence of reduction due to lack of effective heuristic information,a positive domain change was proposed to measure the significance of kernel granularity,and the change of boundary set was used to measure the significance of non-kernel granularity.The new measurement method could measure not only the significance of kernel granularity but also that of non-kernel granularity.Based on the new granularity significance,an improved pessimistic multi-granularity reduction algorithm was proposed.Compared with the heuristic attribute reduction algorithm of sample selection,the fuzzyε-approximate reduction algorithm of information entropy,the reduction algorithm of granularity acceleration and the feature selection algorithm of neighborhood discrimination index,the new algorithm could reduce the number of iterations and found the granularity reduction subset more efficiently.Through the experimental analysisof University of California Irvine(UCI)data sets,it was proved that the algorithm was effective and practical.

关 键 词：多粒度粗糙集 粒度重要度 粒度空间 粒度约简 核粒度 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]