Research On Construction Of Multi-granulation Knowledge Spaces And Attribute Reduction Methods

In recently years,with the rapid development of new information technology such as cloud computing,internet of things,social network and artificial intelligence,the explosive growth of various data has changed the life style of human.Faced with the rapid growth of data,how to quickly and efficiently acquire knowledge from the massive data has become the hotspot in the field of relative research.As a new tool for solving complex problems in different knowledge granulation space,granular computing has been widely used in data mining,fuzzy information processing,large-scale computing,cloud computing and so on.Granular computing mainly includes rough set theory model,fuzzy set theory model and quotient space theory model.And,the correlation research between the different theoretical models mentioned ahead not only provides a solution for the complex knowledge problems,but also improves the theoretical basis for the research of uncertainty knowledge acquisition.Based on the idea of multi-granulation in granular computing,the relative research works are presented as follows:(1)For an uncertainty concept,the different approximate representation can be obtained by changing the information granulation.Firstly,the concept of knowledge granulation of binary relation is put forward for the problem of information granulation representation.And the changing rules of granulation series and approximation accuracy series of an uncertain set is presented for describing a hierarchical quotient space structure.A new utility function which is constructed by knowledge space granulation and approximation accuracy is proposed.For automatically searching optimal knowledge space,an algorithm based on the utility function is put forward.(2)The classification isomorphism and granulation isomorphism between two hierarchical quotient space structures are defined and discussed.Through the hierarchical quotient space structure of a fuzzy equivalence relation,the essence of classification ability of a fuzzy equivalence relation is also revealed.Finally,a new algorithm for establishing isomorphic fuzzy equivalence relations based on a hierarchical quotient space structure is presented,and this algorithm can further explain the nature of those isomorphic fuzzy equivalence relations from a new perspective.(3)From the perspective of multi-granulation,the knowledge measurement method in approximate set of rough sets is studied.Considering the changing of spatial granulation,a new concept,fuzziness of an approximation set of rough sets is put forward firstly.Then the changing rules of fuzziness in changing granulation spaces are analyzed.It is proved that the fuzziness of approximate set of rough sets will show a monotonically decreasing as the knowledge threshold increasing.And,this monotonically changing provided the theory foundation for designing a heuristic knowledge reduction algorithm based on fuzziness of approximate set.Finally,based on the fuzziness of approximation set,a new algorithm for attribute reduction is presented.Several experimental results show that the proposed method has relative better classification characteristics.