A novel weighted combination method of conflicting evidences

Dempster combination rule, a classical combination rule, has many excellent properties, but it involves counter-intuitive behaviors when the evidence highly conflicts with each other. In order to solve this problem, a novel weighted evidence combination method is proposed. By introducing the distance of evidence, the mean square value of the Jousselme's distance is defined as new measure criteria of conflicting evidence, and then the original evidences are divided into two different parts based on new criteria. For the two parts, weights of evidences which represent the importance degrees of evidences are determined by using uncertainty measure, respectively. Finally, the original evidences are modified by the discounting coefficient method. Based on the Dempster's rule of combination, the rational results can be obtained. The proposed method not only possesses the superiority of information convergence like the classical Dempster's rule but also avoids the unreasonable combination results when evidence conflicts with each other. Some numerical examples provided show the efficiency and rationality of the proposed method.