一种改进的证据合成方法

Improved evidence combination approach

D-S证据理论不能很好地描述证据之间的冲突,而且证据高度冲突时合成规则会得出反直观的结果。针对这一问题,提出了一种改进的证据合成方法。首先建立余弦相似度空间,利用证据向量之间的夹角余弦度量证据相似性程度,通过冲突证据检测因子对其进行分类;然后引入冲突比例因子决定证据的修正方法,利用相似度对其进行局部修正或全局修正;最后将修正后的证据代入D-S公式进行合成。应用实例证明,该方法能够判定冲突证据,实现冲突证据和相似性证据的合成,具有较好的稳定性、分类精度和收敛速度。

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 combina- tion approach from the aspect of clustering analysis. Firstly, it built up cosine similarity space, in which cosine between evi- dence 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 exam- ples prove that this approach can differentiate conflict evidence, processes good stability, classification precision and conver- gence speed, so it is suitable for the combination of similarity evidence and conflict evidence.