Evidence Reasoning Method Research Based On The ISODATA Clustering And Improved Similarity Measure

As a method of uncertainty reasoning, D-S evidence theory has been widely used in the field of artificial intelligence, detection and judgment, information fusion, etc. because of the simple and flexible reasoning mechanism. But if there is high conflict between evidence when use evidence fusion formula for evidence fusion, then the result may contradicts with the fact, or even lead to error, which has become the main problem encountered in the practical application of evidence theory. Based on this, this paper conducts research on the evidence reasoning, and put forward a new method of evidence fusion.For the problem that the highly conflictive evidences can not be processed by Dempster rule in evidence reasoning, the reason of conflict occurrence is further revealed by analysis of uncertainty among evidences and evidences themselves. That is to say, the conflict of evidence not only stems from the contradiction between different evidence, but also stems from the uncertainty of evidence themselves. Then a similarity measure is proposed by considering both self-conflict and external conflict of evidence, the credibility of evidence is calculated utilizing the new similarity measure, and the initial evidences is modified according to the credibility. At the same time, according to the clustering characteristic of initial evidences, they are clustered by applying ISODATA algorithm. After that, all evidences in each cluster are combined by applying Dempster rule as the representative of evidence. The reliability of the representative of evidence is calculated by considering both the credibility and the frequency of evidence in a cluster. Finally, all the representatives of evidence are combined by using the unified combination rule. Then, The method proposed in this paper is deeply compared with other methods through massive numerical examples. The results show that the new method proposed in this paper has distinct advantage and high reliability over others and can effectively solve the combination problem of conflictive evidences.