摘要
粒计算理论从多个角度、多个不同的粒度层次出发,对不确定、不精确或复杂的问题进行求解,现已成为人工智能领域研究的一种重要方法。针对决策系统属性约简与高效决策的粒度选择问题,分析了多粒度决策系统中信息粒与粒度划分的概念,定义了粒化度量和粒结构关于对象的粒化粗糙度,能够准确地反映决策系统中不同粒结构下的知识粒度大小。为弥补传统决策系统约简往往只考虑基于论域属性约简的缺陷,讨论了基于对象的局部约简方法,提出了基于论域和对象的决策系统最优粒度选择约简算法,并结合实例验证了该算法的有效性。
Granular computing,as an important theory method of artificial intelligent,studies the solution of uncertain,imprecise issues or complicated problems from different angles and granularity levels.On the basis of decision system theory of multi-granularity,information granulation and granularity partition were analyzed through different granularity levels.Then the concepts of granulating measurement and granular roughness which can exactly express the size of different granularity partition were defined for the problems of attribute reductions and efficient decision making in decision system.After discussing the local reduction method based on objects,an algorithm of optimal granularity reductions was proposed based on both universe and objects for overcoming the drawbacks of decision system reductions in traditional methods,which are only focused on the universe of decision system.Finally,the experimental results show the validity of the proposed algorithm.
出处
《计算机科学》
CSCD
北大核心
2018年第2期152-156,共5页
Computer Science
基金
国家自然科学基金(U1304403)
2017年度河南省高等学校重点科研项目(17B520036)
2016年许昌市科技局基础与前沿计划研究项目:基于单核苷酸多态性位点挖掘的动物种群结构研究资助
关键词
多粒度
最优粒度
决策系统
粒化度量
局部约简
Multi-granularity
Optimal granularity
Decision systems
Granulating measurement
Local reduction
