A novel evidence discounting for conflicting evidences combination

摘要

Dempster's rule of combination is a core part in evidence theory. However, it may generate an unreasonable result when fusing conflicting evidences. For the sake of dealing with such a problem, a novel discounting method is proposed. Minimum singular value of BPA (basic probability assignment) matrix consisting of two BPAs corresponding to two evidences can measure the dissimilarity between these two evidences more accurately than some other dissimilarity measures. Thus, minimum singular value is applied to determine discounting factor. Then, the discounted evidences are fused by Dempster's rule to make the final decision. This presented discounting method can maintain both the diversity of the evidences and the good properties of Dempster's rule associativity and commutativity. Numerical example illustrates that this evidence discounting method presented can efficiently fuse conflicting evidences and has faster convergence speed, better focusing degree than some existing approaches.