证据融合的聚焦与冲突处理研究  

Research on convergence and conflict treatment in evidence fusion

作　　者：李烨[1,2] 王亚刚[1,2] 许晓鸣[1] 

机构地区：[1]上海理工大学光电信息与计算机工程学院,上海200093 [2]上海理工大学工业过程自动化国家工程研究中心,上海200093

出　　处：《系统工程与电子技术》2012年第6期1113-1119,共7页Systems Engineering and Electronics

基　　金：国家自然科学基金(61074087);上海市教育委员会科研创新项目(12ZZ144);上海理工大学光电学院教师创新能力建设项目(GDCX-Y1111)资助课题

摘　　要：针对Dempster组合规则存在的缺陷,即:在融合高度冲突的证据时会产生违反直觉的结论,无法处理完全不一致的证据,而且不能根据命题子集基数向下聚焦,提出了一种结合修改组合规则与修改证据源两种不同冲突处理策略的证据融合方法,采用基于Jousselme证据间距离指标的加权平均法对原始证据进行修正,并给出新的组合规则对修正的证据进行多次组合。与当前代表性方法相比,该方法在根据子集基数向下聚焦、收敛趋势以及冲突证据的处理方面都有好的表现。There are some defects with Dempster＇s rule of combination. It may draw a counter-intuitive conclusion when combining highly conflicting evidences and cannot deal with contradictory evidences. Moreover, it cannot converge according to the cardinality of the sets in the evidences. An evidence fusion method is pro- posed which is an integration of two different conflict treatment strategies, i. e. , modifying the combination rule and modifying the evidences. The original evidences are modified by a weighted average method based on Jous- selme＇s measure of distance between evidences and a new combination rule is proposed to combine the modified evidences for several times. The comparative study with current several representative methods show that the proposed method performs well in converging according to the cardinality of sets, convergence trend and dealing with conflicting evidences.

关 键 词：信息融合 不确定推理 证据理论 证据冲突 聚焦 

分 类 号：O21[理学—概率论与数理统计] TP274[理学—数学]