基于神经网络求解三值模型下的故障诊断问题  

Fault Diagnosis under a Three-Valued PMC Model Implemented by Artificial Neural Network

作　　者：孙丽萍[1] 杨小帆[1] 刘怀义[1] 

机构地区：[1]重庆大学计算机学院,重庆400044

出　　处：《计算机技术与发展》2008年第4期16-18,177,共4页Computer Technology and Development

基　　金：教育部新世纪优秀人才资助计划(NCET-05-0759);教育部博士点基金资助项目(20050611001);重庆市自然科学基金资助项目(CSTC2006BB2231;CSTC2005BB2191)

摘　　要：在多处理机系统的系统级故障诊断中,一个重要的研究课题是确定最可能故障处理机集,该问题可以归结为NP-完全的整数线性规划问题。连续Hopfield神经网络能够近似求解最优化问题,因此是解决这类问题的可选路径。文中主要研究如何构建连续Hopfield神经网络,以在三值PMC模型下近似地确定最可能故障集,相比于常用的二值诊断模型,能得到更准确的诊断结果。在超立方体结构上进行了一系列的数值实验,仿真结果表明:该方法具有实用性。In the context of system- level fault diagnosis of multiprocessor systems,one major issue is to identify a set of processors that are most probable to fail. This problem can be reduced to the NP-complete integer linear programming problem. It is well known that continuous Hopfield neural network can be employed to solve optimization problem. Studies how to construct Hopfield neural network to identify the sets of faulty nodes under a three - valued PMC model. In comparison with the two-valued diagnosis models, it can get more accurate diagnosis results. Finally,series of numerical experiments were done on hypercube,and simulation results justify the method.

关 键 词：系统级故障诊断 Hopfidd神经网络 三值PMC模型 超立方体 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] TP306.3[自动化与计算机技术—控制科学与工程]