摘要
多粒度是当前粒计算研究的一个重要方面.在实践中,人们往往选择比较合适的粒度层次来解决问题.作为信息系统的一种特殊情况,多粒度决策系统是经常使用数据表示形式.在这样的系统中,对象可以在属性的不同粒度层次上取不同的观测值.实际使用时,常常遇到在数据属性上需要比较大小,即属性带有序关系.序关系分析是多指标决策的重要内容,而粗糙集是一种处理序关系有效方法.围绕多标记序决策系统的知识获取问题来开展研究,首先,介绍了多标记序决策系统的概念;然后,在协调的多标记序决策系统中定义了最优粒度和局部最优粒度,并介绍了基于局部最优粒度的属性约简和规则获取方法;最后,在不协调的多标记序决策系统中引入了广义决策,定义了广义最优粒度和广义局部最优粒度,并给出了基于广义局部最优粒度的属性约简和规则获取方法.
Multi granularity is an important issue of granular computing.In practice,people tend to choose the appropriate level of granularity to solve problems.As a special case of information system,multi-granular decision system can usually be observed in real-life world.In such system,objects may take different values under the same attribute measured at different granularity levels.When used in practice,it is often encountered that the data attribute needs to be ordered,that is,the attribute has an ordered relationship.Ordered relationship analysis is a class of important issues in multi-criteria decision making.Rough set is an effective approach to handle ordered relationship.Due to the rampant existence of multi-granular ordered decision systems in real world,the purpose of this study is to select appropriate level of granularity and to discuss rule acquisition from multi-granular ordered decision systems.Aiming at rules acquisition in multi-label ordered decision systems,the concept of multi-label ordered decision systems is introduced firstly.Then,notions of the optimal granularity and the local optimal granularity in consistent multi-label ordered decision system are defined,and the approaches to attribute reduction and rule extraction based on the local op-timal granularities are explored.Finally,the generalized decisions are introduced to inconsistent multi-label ordered decision systems,the generalized optimal granularity and the generalized local optimal granularity are also defined,and the approaches to attribute reduction and rule extraction based on the generalized local optimal granularities are further investigated.
出处
《南京大学学报（自然科学版）》
CAS
CSCD
北大核心
2017年第6期1012-1022,共11页
Journal of Nanjing University（Natural Science）
基金
国家自然科学基金(61602415
61573321
61272021
41631179)
浙江省海洋科学重中之重学科开放基金(20160102)
关键词
粒计算
多标记
序决策系统
局部最优粒度
granular computing, multi-label, ordered decision system, local optimal granularity
