面向多源信息冲突管理的自适应折扣系数生成

Eliciting Adaptive Discounting Coefficient for Multi-source Conflict Management

在应用证据理论融合多源信息时,信任函数冲突是面临的一个重要难题.基于闭世界假设,冲突的产生是由于信息源的误判或无知,对不可靠信息源采用折扣系数弱化其作用是一个解决途径,然而折扣系数生成是一个被忽略的问题.为此,本文提出一种基于证据可信度的自适应折扣系数生成方法,根据证据之间的相似性导出合理的证据可信度,并根据该可信度与提供证据的信息源可靠性共同生成折扣系数,信息源的可靠性通过后验对比进行更新.多个实例结果表明,该方法能对多信息源的一致或冲突证据进行有效合成,且合成结果与人们的直觉一致.

Within the framework of evidence theory,conflicts between belief functions have been an intractable problem when fusing information provided by multi-sources.Based on the assumption of closed world,the appearance of conflicts stems from misjudgment or ignorance of sources.An alternative solution is applying ＂discounting coeffient＂ method to discount unreliable sources so as to weaken their effects.Nevertheless,the generation of discounting coeffient has often been neglected.Therefore,a method for eliciting adaptive discounting coeffient based on evidence credibility is presented in this paper.The rational credibility of evidences can be derived according to the similarity between evidences,and together with reliability of sources,adaptive discounting coeffients are then acquired.The reliability of sources can be updated later by comparison with posterior estimates.Several numerical examples illustrate that the proposed method can work efficiently with compatible or conflict evidences of multi-sources,and the results of combination are also consistent with people＇s intuition.