基于二进神经网络的0/1分布系统可靠性分析  

Reliability of Systems with 0/1 Distribution Based on Binary Neural Networks

作　　者：杨娟[1,2] 陆阳[1] 黄镇谨[1,3] 

机构地区：[1]合肥工业大学计算机与信息学院,合肥230009 [2]合肥工业大学信息与通信工程博士后科研流动站,合肥230009 [3]广西工学院计算机工程系,柳州545006

出　　处：《自动化学报》2014年第7期1472-1480,共9页Acta Automatica Sinica

基　　金：安徽省自然科学基金项目(1408085QF117);合肥工业大学博士专项科研资助基金(2013HGBZ0182);合肥工业大学青年教师创新项目(2013HGQC0019)资助~~

摘　　要：系统可靠性的计算依赖于各基本单元的0/1分布关系及其构成的布尔逻辑.本文利用二进神经网络可以完备实现布尔逻辑的特性,提出一种基于二进神经网络的可靠性分析方法.该方法针对每个二进神经元的输入都是0/1逻辑关系的线性组合这一特点,提出并且证明了0/1分布的线性组合的概率分布函数;建立系统功能与布尔函数问的等价关系,将系统转化为相应的二进神经网络;利用线性组合的概率分布函数,通过逐层计算该二进神经网络的0/1输出概率,解决了一股系统的可靠性计算问题.The computing of system reliability relies on the relationship of 0/1 distribution of components and their boolean logic. With the help of the characteristic that binary neural networks can complete the whole boolean function, we propose a method of reliability analysis based on binary neural networks. According to the input of every binary neuron is a 0 or 1 logic variable, we provide and prove the distribution function of the linear combination of 0/1 distribution. Then the equivalent relation between the system function and boolean function is established, and the system is converted to an equivalent binary neural network. As a result, using the distribution function, we can successfully resolve the problem of reliability analysis of general systems by computing the 0/1 output probability layer by layer.

关 键 词：二进神经网络 系统可靠性 分布函数 线性组合 

分 类 号：TB114.3[理学—概率论与数理统计]