Research On A DS Evidence Information Fusion Method For Divergence Measure

Because DS evidence theory has the advantage of dealing with uncertain and highly conflicting evidence,DS evidence theory is widely used in dealing with multi-source information fusion.Although divergence has made some achievements to measure the difference of DS evidence in the application of dealing with highly conflicting evidence,there is still an open subject on divergence research.This paper constructs a new divergence measure and applies it to target identification and fault diagnosis.The main contents are as follows:A novel divergence measure is constructed.The form of the divergence is the integral of the likelihood value of the proposition as the upper limit and the reliability value of the proposition as the lower limit.Furthermore,it is theoretically proved that the new divergence satisfies boundedness,non-degradation,and symmetry.The three properties are visually demonstrated by numerical simulation.The newly proposed divergence is an extension of the classical BJS divergence.If the proposition is a single element set,the new divergence degenerates to JS divergence.If the intersection of all propositions of two bodies of evidence is empty,the new divergence degenerates to BJS divergence.Compared with the two classical divergences,the novel divergence in this paper is superior illustrated by the empirical research,and the conclusion can quantify the difference of evidence bodies,which is also in line with our intuitive judgment.Based on the novel divergence,an information fusion model is constructed.First,the credibility weight of the evidence body is obtained by using the new divergence measure.Secondly,the information volume weight of the evidence body is obtained by Deng entropy.Finally,the comprehensive weight of the evidence body is obtained based on the credibility weight and the information volume weight.Thus,the mass of the proposition is modified by comprehensive weight.Then,Dempster’s combinationrule is used for fusion.This model was applied to two practical problems of target recognition and fault diagnosis,respectively.Compared with Dempster’s method,the BJS-based method,and other methods,the results showed that the information fusion model based on the novel divergence improved the accuracy of target recognition andthe accuracy of fault diagnosis.This paper provides a different way to deal with high-conflict evidence fusion,and enriches the theory of information fusion.