New method for measuring the degree of conflict among general basic probability assignments
New method for measuring the degree of conflict among general basic probability assignments
复制标题
测量一般基本概率分配之间冲突程度的新方法
DOI：
10.1007/s11432-011-4346-0
复制
发表时间：
2012-02-01
影响因子：
8.8
通讯作者：
Han DeQiang
中科院分区：
文献类型：
--
作者：
He You;Hu LiFang;Han DeQiang
在智能推理中，由于处理冲突信息时会出现反直觉的结果，使得Dempster组合规则受到质疑。因此，在多源信息融合与不确定性推理的文献中，提出了许多改进的组合规则。然而，证据冲突的识别问题一直被忽视。定义了一个新的冲突证据度量参数--冲突距离参数，通过对现有冲突参数的分析，判断证据中是否存在冲突。同时，提出了合理的冲突测度函数应满足的判定准则。最后，对两种典型的反直觉情形进行了分析，表明新参数不仅能满足判定规则，而且在判定是否使用Dempster组合规则时能达到与二维测度相同的效果，在判定是否使用Dezert-Smarandache理论时能超过二维测度的效果.
There is doubt about Dempster's rule of combination because of counter-intuitive results when dealing with conflict information in intelligent reasoning. Therefore, many modified combination rules have been presented in the literature on multi-source information fusion and reasoning with uncertainty. However, the issue of identifying conflict among evidence has been ignored. A new parameter measuring conflict evidence called the conflict distance parameter is defined to determine whether there are conflicts in evidence based on the analysis of existing conflict parameters. At the same time, a decisive rule, which a reasonable conflict measure function should satisfy, is put forward. Finally, an analysis of two typical counter-intuitive situations is given, showing that the new parameter can not only satisfy the decisive rule, but can also attain the same effect as a two-dimensional measure when deciding whether to use the Dempster's rule of combination, and can exceed its effect when deciding whether to use Dezert-Smarandache theory.