Research On Granular Computing Method For Knowledge Acquisition In Lattice-valued Information Systems

As an important tool for knowledge acquisition and data mining,granular computing is a novel approach to simulate human nature thinking patterns in solving large-scale complex problems.The basic idea of this new theory is to describe specific problems and reasoning by using information granules from different views and different levels in the process of solving problems.This thesis mainly studies granular computing theories and knowledge acquisition method in lattice-valued information systems based on the rough set theory.This research field covers the basic mathematics,applied mathematics and information science.These results can enrich theoretical research in lattice-valued information systems,and can propose a new method for dealing with practical problems such as decision management,market forecast,medical diagnosis,and so on.Results and innovations are as follows.(1)The mathematical structure and representation theories of information granularities are explored in lattice-valued information systems.From the view of algebra,we firstly propose knowledge vectors,knowledge spaces,and knowledge granular levels.Secondly,operation rules are considered in order to describe the knowledge with different standards.Moreover,we obtain an important result.That is to say,the position of each knowledge vector is demonstrably determined.In other word,a knowledge vector is only in a certain level,and the same granular lever has some different knowledge granules.(2)Some uncertainty measures are discussed by introducing information entropy,knowledge resolution and rough entropy in lattice-valued information systems.Some important properties are studied,and the relations are considered carefully among them.It has proved that the rough entropy and knowledge granularity are monotonically decreasing as the knowledge certainty degree grows.Moreover,it can be shown that rough entropy of rough set can accurately characterize the rough set rough degree.These conclusions may further polish the theoretical basis of lattice-valued information system.(3)Multi-granularity Rough Set model are established in lattice-valued information systems.Classical Rough set model is based on equivalence relations.However,this model cannot deal with multiple complex granularities in lattice-valued information systems.Therefore,optimistic and pessimistic multi-granularity Rough Set models are proposed in lattice-valued information systems.We investigate the important properties of them,and study the relationship among optimistic,pessimistic multi-granularity Rough Set and single-granularity Rough sets.Furthermore,roughness and dependence are proposed in two kinds of multi-granularity Rough set model.These concepts and properties are important foundation of multi-granularity Rough set model in lattice-valued information system.(4)A novel method of attribute reductions and rules acquisition is obtained in lattice-valued information systems with a fuzzy decision.A rank approach with dominance classes is considered by proposing dominance degree in the complex systems.Moreover,approximation reductions are constructed in lattice-valued information systems with a fuzzy decision,and dominance rules acquisition is also discussed in this system.Furthermore,two algorithm analyses to attribute reduction are developed in information systems with a fuzzy decision,and seven real-life datasets are calculated by the computer program.From the experiment results,it is illustrated that how to make a decision by using the proposed approach.