The Exploration And Application Of Belief Function Approximation

D-S evidence theory is a useful uncertainty reasoning theory, which haves an advantage of the expression and fusion for uncertainty, and it has been successfully applied to data fusion, target recognition and intelligent decision-making etc. When applied to practical project, D-S evidence theory often encounters all kinds of conflict evidence. D-S evidence combination rules may come to the conclusion contradicting with the fact, when there is high degree of conflict between evidences. Therefore, it limits the wide application of D-S evidence theory. In addition, it also limits the application of evidence theory in the practical for its large calculation, which mainly caused by the computational complexity that increasing exponentially with the growth of the cardinality of recognition framework. Therefore, our main research is about optimizing the fusion of conflict evidence and the belief function approximation.First of all, the paper introduces the background and the significance of the evidence fusion and belief function approximation, after which we sketch the research status of evidence theory and its basic knowledge.Secondly, it introduces several improved methods of evidence fusion, which aims to solving the shortages of D-S evidence combination rule when dealing with conflict evidence. Then, we give a general combination rule based on the distribution of local conflicts after analyzing their advantages and disadvantages. In addition, the paper puts forward a new improved method of evidence combination based on the distribution of local conflicts, and example shows the effectiveness of this method.Then, the paper introduces the definitions and theorems of belief function approximation. Aiming to its disadvantages, we proposes a method of belief function approximation based on decomposing focal elements with big cardinality by analyzing the existing method of belief function approximation.Finally, the paper proposes a method of belief function approximation based on separable mass function by the properties and theorems of separable mass function.