摘要
潜在问题是影响大型复杂系统安全性、可靠性的重要因素.神经网络是一种新的潜在问题分析方法,但是其分析结果难以解释.本文提出了一种基于电路结构的神经网络模型(Neural network model based on circuit architecture,CArNN),将CArNN作为个体进行集成,形成神经网络集成用于潜在问题分析.对CArNN模型的鲁棒性进行了分析,提出了两个保证模型鲁棒性的约束条件.利用此方法对一个经典电路进行了分析,结果显示,此方法对潜在电路的正确识别率达到94%.
Sneak circuits have important influence on systems' safety and reliability. Artificial neural network (ANN) is a new sneak circuit analysis (SCA) method. However, it is difficult to interpret why ANN gives the conclusion. A novel coadjacent neural network model based on circuit architecture, named CArNN, is proposed. Ensemble of the CArNN was employed to analysis the sneak circuits existing in systems. Two constraints guaranteeing the robustness of CArNN were given. A typical circuit containing sneak circuit was used to verify the method. The results showed 94% of the sneak circuits can be correctly recognized by the method.
出处
《自动化学报》
EI
CSCD
北大核心
2008年第2期188-194,共7页
Acta Automatica Sinica
基金
国家自然科学基金(60736026)
教育部新世纪优秀人才支持计划资助~~
关键词
潜在问题分析
可靠控制
神经网络集成
克隆选择算法
鲁棒性分析
泛化性能
Sneak circuit analysis, reliability control, neural network ensemble, clonal selection algorithm, robustness analysis, generalization performance