【摘要】 研究了虚拟环境下基于知识融合的多Agent仿真决策中的冲突证据合成,认为传统的Dempster-Shafer证据组合规则导致悖论的主要因素是归一化因子,使结果产生了误差。对该规则进行了形式变换,分解了归一化因子,把Dempster-Shafer分为两项之合,其中第二项含有未知项。文中首次提出了融合误差的概念,引入误差函数项和全局平均支持度,这两项与未知项相关。采用等效的方法,提出了一种改进的证据组合规则,与变换后的Dempster-Shafer规则形式上一致,仅把误差赋给了未知项。理论分析和实验表明,新的合成规则得到了合理的融合结果,产生的误差最小。最后给出了一个高速贴标机的多Agent仿真决策的实例说明新组合规则的有效应用。更多还原

【Abstract】 The combination of conflicting evidences of multi-Agent decision was studied based on knowledge fusion in virtual environment.The error,unanti-result was caused by normalized constant in the Dempster-Shafer theory.The combination of evidence in Dempster-Shafer theory was transformed formally that the formula was disassembled into two terms and the normalized constant was also disassembled.The second term contains an unknown.So a new improved combination rule with equivalent method was proposed and the concept of the errors of the fusion was also proposed.Then the errors of function and average of conflicting evidences were formulated,and they are correlative with the unknown term.The new approach is consistent with transformed Dempster-Shafer’s rule in the form.Only the errors were assigned to the unknown term in the new approach.The experimental results and theoretical analysis show that the new methodology can produce reasonable results and its error of the fusion is minimum.Finally,an example of multi-Agent simulation decision of labeler was illustrated that the new approach is effective and reliable.更多还原