Improved evidence combination approach
Guiming Chen · Jisuanji yingyong yanjiu · 2013
In D-S evidence theory,conflict coefficient cannot well depict the conflict between evidence,and counterintuitive results generate in the combination of highly conflicting evidence.Aiming at this question,this paper proposed a new combination approach from the aspect of clustering analysis.Firstly,it built up cosine similarity space,in which cosine between evidence vectors was used to measure the similarity degree,and then classified the evidence by way of conflict evidence detection coefficient.Moreover,it introduced conflict proportion coefficient to decide the modified method—local modification or full modification according to its similarity.Finally,it input the modified evidence to D-S combination formula.Application examples prove that this approach can differentiate conflict evidence,processes good stability,classification precision and convergence speed,so it is suitable for the combination of similarity evidence and conflict evidence.