采用聚类分析的冲突证据判定及合成方法

针对D-S 证据理论中高度冲突证据合成时产生反直观结果的问题，从聚类分析的角度提出了一种新的冲突证据合成方法。首先建立余弦相似度空间，利用证据向量之间的夹角余弦定义冲突证据检测因子，进而提出冲突证据判据判定冲突证据，从而实现证据分类；其次引入冲突比例因子决定冲突证据的修正方式，通过绝对相似度对冲突证据进行局部修正，通过相对相似度对证据进行全局修正；最后将修正后的证据带入D-S 公式进行合成。应用实例证明:新方法能够判定并有效合成冲突证据，具有较好的分类精度、收敛速度和稳定性。

Approach of conflict evidence judgment and combination based on clustering analysis

D-S evidence theory cannot combine highly conflict evidence and brings counterintuitive results. In order to solve this issue, a new conflict evidence combination approach was proposed from clustering analysis. Firstly, cosine similarity space in which evidence was regarded as space vecter was built up, and cosine between evidence vectors was calculated to define conflict evidence detection coefficient. Based on the space and coefficient, conflict evidence criterion was presented to judge the conflict evidence, and the evidence was classified into similar evidence and conflict evidecne. Secondly, conflict proportion coefficient was introduced to decide the modified method in which conflict evidence was modified by use of the absolute similarity and all the evidence was modified by use of the relative similarity. Finally, the modified evidence was put into D-S combination formula to combine. Application examples prove that this new method can judge conflict evidence and combine them effectively, and processes good classification precision, convergence speed and stability.