The Combination Method of Conflict Evidence Based on Classification Correction
Kehua Yang, Yuping Feng · 2016
In order to efficiently combine highly conflict evidence, a new method of evidence combination was presented. Firstly, since traditional evidence distance can't measure the degree of similarity between evidence effectively in some cases, the paper analyzed traditional evidence distance functions and combined with Jousselme distance, then presented a new evidence distance. And then, reliability of each evidence was obtained to determine relative reliability and weight factors, according to the new evidence distance. Secondly, evidence was classified into three categories: consistent evidence, non-conflict evidence and conflict evidence, according to the new evidence distance parameter and the local conflict parameter. Finally, different evidence was revised in different ways by using relative reliability and weight factors, then revised evidence was combined by means of Dempster's combination rule. A numerical example shows that the proposed method can combine highly conflict evidence efficiently and accelerates convergence.