Research On Fuzzy Rule Extraction And High-conflict Evidence Fusion Based On D-S Evidence Theory

Fuzzy inference system has the ability to process fuzzy information,and has been widely used in pattern recognition,positioning,fault diagnosis and so on.Fuzzy rule base is the core of a fuzzy inference system,and data-driven fuzzy rule extraction is an important research topic.Conflicting rules are usually extracted during the rule extraction process.For a group of conflicting rules,how to choose the rule reasonably and build the rule base is a meaningful research topic.In the field of multi-source information fusion,Dempster-Shafer theory(DST)is an effective tool.However,counter-intuitive results are produced when the Dempster’s combination rule is applied to fuse multiple highly conflict evidences.How to realize the effective fusion of highly conflict evidence and avoid the generation of counter-intuitive results has become an important issue.In view of the two problems mentioned above,this thesis has done the following work:(1)A fuzzy rule extraction method based on DST is proposed.Each piece of data is regarded as an evidence,and all the candidate rule consequents constitute the frame of discernment.Firstly,the basic probability assignment(BPA)functions corresponding to evidences are constructed based on the similarity measure between fuzzy numbers.Then,evidences which are in the same fuzzy region are fused according to Dempster’s combination rule.After that,a new BPA which reflects the fused information can be obtained.Finally,the consequent of the fuzzy rule can be determined according to the fusion result.In addition,an optimization strategy for centers of output fuzzy sets for rules based on BPAs is developed.Experimental results illustrate that both the proposed fuzzy rule extraction method and the optimization strategy have better performance compared with others.(2)Following the idea of modifying evidence sources,two different weighted evidence combination approaches are proposed.The first approach based on multi-criteria strategies.It evaluates evidences comprehensively by considering multiple different measurement perspectives.Firstly,the mutual support degree between evidences is determined according to the evidence distance,evidence angle and the difference between evidences.On this basis,the comprehensive credibility of the evidence is determined by combining the support degree and the discriminability of evidence itself.Thereby,the weight of each evidence can be derived.The second approach introduces the simplex into the internal conflict measurement to measure the discriminability of evidences.The support degree of evidences can be determined based on the evidence distance and the angle between the line from the evidence to the center of the simplex.The discriminability of evidences is evaluated based on the distance between the evidence and the center of the simplex.Combine the two measures mentioned above,the credibility of evidences can be obtained,and then the value of weights are determined.Numerical results indicate that the proposed two combination methods can not only fuse highly conflict evidences effectively,but also have better focusing degree than other existing approaches.