模糊决策表中基于OWA算子的三支属性约简  

Three-Way Attribute Reductions Based on OWA Operator in Fuzzy Decision Table

作　　者：杨霁琳[1] 张贤勇[2] 唐孝[2] Yang Jilin;Zhang Xianyong;Tang Xiao(College of Fundamental Education,Sichuan Normal University,Chengdu,610068,China;College of Mathematics and Software Science,Sichuan Normal University,Chengdu,610068,China)

机构地区：[1]四川师范大学基础教学学院,成都610068 [2]四川师范大学数学与软件科学学院,成都610068

出　　处：《数据采集与处理》2018年第4期712-721,共10页Journal of Data Acquisition and Processing

基　　金：国家自然科学基金(61673285;61203285;61303204)资助项目;四川省青年科技基金(2017JQ0046)资助项目;四川省教育厅科研(17ZB0356)资助项目

摘　　要：在模糊决策表中,基于有序加权平均(Ordered weighted averaging,OWA)算子建立的相容关系中,OWA算子是一个信息集成工具。本文利用OWA算子及其诱导的截集相容关系,分别讨论基于正域、负域与边界域的属性约简。首先,利用OWA算子中的权重给出属性区分对象的贡献度;然后,根据属性贡献度定义每个属性被约简的可能性;从而给出模糊决策表的一种启发式三支属性约简方法,该方法可以减少属性约简的搜索空间并避免属性约简的盲目性;最后,利用实例来分析影响该三支区域属性约简方法的主要因素,并说明该启发式属性约简方法的合理性和有效性。本文提出的属性贡献度度量及启发式三支属性约简方法将减少属性约简的搜索空间,有益于模糊决策表特征选取的应用。In the fuzzy decision table,the tolerance relation is built based on the ordered weighted averaging(OWA)operat orand,and OWA operator is a basic tool of information integration.Based on the OWA operator and its induced tolerance relation of cut sets,attribute reductions of th e positive region,negative region and boundary region are discussed.Firstly,an attribute contribution degree distinguishing objects is defined by the weight of the OWA operator,to describe the reduction possibility of attributes.Conseq uently,a method of three-way attribute reduction based on the reduction possibility is proposed in the fuzzy decision table.The method reduces the search space and avoids reduction blindness.Finally,an example is analyzed to show the in fluencing causes of attribute reductions and the validity of the proposed method.The proposed measurement method of attribute contribution degree and the heuristic method of attribute reduction based on the three regions will reduce the searching space of attribute reduction,and thus they are benefit for applications of the fuzzy decision table.

关 键 词：三支决策 粗糙集 属性约简 OWA算子 相容关系 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]