基于环境独立机制的静态离散贝叶斯网络参数简化方法  

Reductive Method of Static and Discrete Bayesian Networks' Parameters Based on Context-specific Independence

作　　者：刘琳颖 孙秦[1] 王瑶[1] 

机构地区：[1]西北工业大学航空学院,西安710072

出　　处：《航空工程进展》2014年第3期332-337,共6页Advances in Aeronautical Science and Engineering

基　　金：国防科工局"十二五"可靠性工程基础预研项目

摘　　要：在大型工程系统的应用中,贝叶斯网络节点参数存储量随系统复杂度呈指数增长。针对该问题,综合飞行器故障模式通常采用多输入单输出逻辑门表达,以及故障树具有特殊静态离散贝叶斯网络的特点,提出将环境独立简化方法应用于常用典型逻辑门的贝叶斯网络。通过基本定理证明利用基于环境独立机制的概率树代替传统的条件概率表,可将呈指数增长的条件概率表减少为线性变化,并定量给出简化后节点参数个数的计算表达式。最后,将该方法用于某飞机前轮转弯系统故障诊断模型,通过与节点原条件概率表的比较,表明概率树方法可有效减少计算机内存需要。In the application of Bayesian Networks（BNs） for large scale engineering systems ,computer memory for the network’s parameters increases exponentially with the nodes of the system .Aiming at this problem and in view of that aircraft failure is often expressed in a multi-input single-output logic gates manner ,as well as that fault tree is a special kind of static and discrete BNs ,the reductive method based on Context-Specific Independ-ence（CSI） is proposed for the application of the BNs modeled from the commonly used typical logic gates .Based on the fundamental theorem ,it proves that when the traditional Conditional Probability Table（CPT） is replaced by the Probability Tree（PT） based on CSI ,the storage need using PT is linear with the nodes ,rather than ex-ponential .Simultaneously ,the formula for calculating the number of parameters needed in a PT is given .Final-ly ,this new method（PT） is used to the fault diagnostic model of an aircraft nose wheel steering system ,in com-parison with the original CPT method ,which shows that PT approach can effectively reduce computer memory needs .

关 键 词：贝叶斯网络 环境独立 概率树 条件概率表 逻辑门 

分 类 号：V37[航空宇航科学与技术—航空宇航推进理论与工程]