基于偏差的冲突证据融合的改进方法

在Dempster-Shafer(D-S)证据理论中,传统D-S组合规则在证据高度冲突时存在失效问题,会产生与常理相悖的结论.为此,文中提出了一种改进的处理冲突证据的融合方法,即通过计算各证据到命题平均支持度的偏差,来检测和消除冲突证据,并对最后结果进行修正,以合成来自不同识别框架的证据.计算实验结果表明,该方法能有效处理冲突,得到符合实际的组合结果,相对于其他典型算法,该方法在收敛性和可靠性上有明显改进.

Improvement of Fusion of Conflicting Evidence Based on Deviation

The paper analyzes the invalidation problem of D-S evidence theory in combining conflicting evidence, and proposes an improved method dealing with the fusion of conflicting evidence based on previous research. The method detects and eliminates the conflicting evidence by calculating the deviation from all evidence to the average support-level for the proposition. And it also amends the end result to handle the combination of evidence from different recognition framework. The numerical experiment shows that accurate identification result can be achieved by this method when the evidence seriously conflicts. Compared with other typical algorithms, the method in this paper improves the reliability and convergence of combination result.