协调广义决策多尺度序信息系统的知识获取  

Knowledge Acquisition for Consistent Generalized Decision Multi-scale Ordered Information Systems

作　　者：张嘉茹 吴伟志 杨烨 ZHANG Jiaru;WU Weizhi;YANG Ye(School of Information Engineering,Zhejiang Ocean University,Zhoushan 316022;Key Laboratory of Oceanographic Big Data Mining and Application of Zhejiang Province,Zhejiang Ocean University,Zhoushan 316022)

机构地区：[1]浙江海洋大学信息工程学院,舟山316022 [2]浙江海洋大学浙江省海洋大数据挖掘与应用重点实验室,舟山316022

出　　处：《模式识别与人工智能》2022年第9期789-804,共16页Pattern Recognition and Artificial Intelligence

基　　金：国家自然科学基金项目(No.61976194,62076221)资助。

摘　　要：决策多尺度信息系统是一类特殊的数据集,系统中的每个对象无论在条件属性集上还是在决策属性上都可取多个尺度的标记值,并且从细粒度标记属性值到粗粒度标记属性值有一个信息粒度变换.文中针对广义决策多尺度序信息系统的知识获取问题展开研究.首先,引入尺度选择概念,一个尺度选择对应一个单尺度的序决策系统,并将优势关系引入广义决策多尺度信息系统,给出在不同尺度选择下对象集的优势类和集合的下近似和上近似的定义及其性质.然后,在协调广义决策多尺度序信息系统中定义5种最优尺度选择的概念,证明实际上只有2种不同类型的最优尺度选择,即最优尺度选择、下近似最优尺度选择、信任最优尺度选择是等价的,而上近似最优尺度选择与似然最优尺度也是等价的.最后,给出协调广义决策多尺度序信息系统的辨识矩阵约简方法,并在最优尺度选择基础上给出蕴含在协调广义决策多尺度序信息系统中的序决策规则.A decision multi-scale information system is a special type of multi-scale data set and each object under each attribute in the system from either the condition attribute set or the decision attribute is represented by different scales at different levels of the granulations,holding a granular information transformation from a finer to a coarser labelled value.To solve the problem of knowledge acquisition in generalized decision multi-scale ordered information systems,the concept of scale selection is firstly defined.Each scale selection is linked with a single-scale ordered decision system.Dominance relations are also introduced into decision multi-scale information systems,representations of information granules as well as lower and upper approximations of sets under different scale selections are presented,and their relationships are examined.Then,five types of optimal scale selections in consistent generalized decision multi-scale ordered information systems are defined.It is proved that there are indeed two different types of optimal scale selections.The notions of optimal scale selection,lower approximation optimal scale selection and belief optimal scale selection are all equivalent,and a scale selection is upper approximation optimal if and only if it is plausibly optimal.Finally,based on optimal scale selections,a method of discernibility matrix attribute reduction and ordered decision rules hidden in consistent generalized decision multi-scale ordered information systems are explored.

关 键 词：多尺度序信息系统 尺度选择 粗糙集 粒计算 知识获取 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]