A Clustering-Based Generalized Evidence Theory Model And Its Application Under Open World

Dempster-Shafer evidence theory(D-S evidence theory)is a method widely used in the field of information fusion,which is used in modeling,representing and reasoning uncertain information.However,the classical Dempster’s combination rule in evidence theory is controversial because of its poor robustness on highly conflicting evidence.Especially,the classical evidence theory is no longer applicable when the framework of discernment is uncomplete.As an expansion of D-S evidence theory,generalized evidence theory can effectively handle the basic problems such as target recognition and combination of evidence in highly conflicting environment.But the research of generalized evidence theory mainly focuses on conflict management,and the research on practical application of information processing under open world is not much.This thesis systematically expounds the developing process and researching status of information fusion and evidence theory firstly,and then briefly introduces the related concepts and definitions of D-S evidence theory and generalized evidence theory.On the basis,this thesis will focus on the exploration and discussion of three basic issues of generalized evidence theory.The first problem is that the existing strong constraint generation method based on fuzzy number can cause loss of original information.A new method is proposed in this thesis,which can generate non-nested belief assignment model without information loss.This method can assign belief to the empty set directly,which is more compatible to the generalized evidence theory.In addition,a comparasion between the proposed method and the existing methods proves the superiority of the former.The second problem is that generalized combination rule has poor performance in combining multiple evidence and cannot model the open world effectively.This thesis analyzes the process of evidence combination,clarifies the causes and significance of conflict,and proposes a new generalized combination rule which can be used in multiple evidence combination.Then,this thesis gives a method to identify unknown targets and complete the framement of discernment.Meanwhile,a hierarchical clustering algorithm which can use the prior information to improve the performance is proposed.Further,the exploration is been done to find the methods to determine the correct number of types.Finally,a introduction of decision-making methods based on evidence theory is given,which can be used to build a complete process of information processing.For each issue,several typical experiments with the analysis of their results will be shown to illustrate the effectiveness of the proposed method.The experimental results indicate that the proposed methods of information processing and decision-making by generalized evidence theory in this thesis is feasible.As the main research conclusion,this thesis obtains a general model and framework of information processing under open world based on generalized evidence theory.