因果图推理的一种新方法  

A New Reason Method in Causality Diagram

作　　者：樊兴华[1] 仲昕[2] 张勤 黄席樾[1] 

机构地区：[1]重庆大学自动化学院 [2]上海交通大学机器人研究所 [3]重庆市科委

出　　处：《计算机科学》2001年第11期48-52,43,共6页Computer Science

基　　金：博士点基金;重庆市科委攻关项目<面向工业应用的智能开发平台及系统研究>资助

摘　　要：1 引言 基于概率论推理的不确定性知识表达推理方法包括信度网[2]、马尔可夫网[2]以及PROSPECTOR[5]中使用的方法等.其中,信度网推理模型因其理论上的严格性和一致性,以及有效的局部计算机制和直观的图形化知识表达,正日益受到高度的重视.然而,信度网也存在一些不足:如处理多连通问题和因果循环问题的方法复杂,计算量大;采用条件概率表达因果关系强度不直观,数据之间存在相依性;较难根据实时收到的信息对知识库中的数据和因果结构进行在线修改;没有考虑条件概率随时间动态变化等问题.Among the probabilistic approaches under uncertainty,belief network is the most representative. The causality methodology which is based on belief network overcomes some difficulties in knowledge express and reason of belief network. So it is very useful for industry application. To improve the deficiency of logic operation complexity and computation complexity,a new reason method in causality diagram has been presented. This method translates causality diagram into some causality trees by normalization and standardization. A new computation method has been brought forth by using the cut sets matrix in former non-intersect causality according to the former non-intersect mind. It can lower the 'NP' difficulty and raise the computation velocity in causality diagram reason. So it is very useful to experiment application such as fault diagnose.

关 键 词：因果图推理 知识表达 知识库 概率 数据结构 

分 类 号：TP311.12[自动化与计算机技术—计算机软件与理论] TP18[自动化与计算机技术—计算机科学与技术]