证据推理的鲁棒性研究  

SOME RESEARCH ON ROBUSTNESS OF EVIDENCE THEORY

作　　者：潘泉[1] 张山鹰[1] 程咏梅[1] 张洪才[1] 

机构地区：[1]西北工业大学自动控制系,西安710072

出　　处：《自动化学报》2001年第6期798-805,共8页Acta Automatica Sinica

基　　金：国家重点基础研究发展规划课题;国家自然科学基金;国防预研科学基金资助课题

摘　　要：在处理不确定信息的方法中 ,证据推理性能突出、应用广泛 ,它的鲁棒性分析十分必要 .首先基于证据推理的目的和要求 ,给出证据推理鲁棒性的定义 ,并提出冲突率的新概念 ,以此为尺度对 Dempster规则、加权分配冲突法和吸收法的鲁棒性进行了深入分析 ,证明了各方法鲁棒性结果 .仿真结果表明改进的组合规则鲁棒性能得到了明显增强 .Evidence reasoning has good performance in dealing with uncertain information and is used in many fields. It is necessary to analyze its robustness. We first give the definition of robustness based on the aim of evidential theory and define a new concept conflict rate. Using the sectional conflict rate, we analyze the robust area of Dempster rule, weigh distribution and absorptive method. Finally, computer simulation indicates that two new rules are more robust than Dempster rule. Moreover, we give a robust area of Dempster rule according to general identification frame.

关 键 词：证据推理 冲突率 鲁棒性 信息融合 不确定推理 人工智能 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]