基于神经网络求解三值模型下的故障诊断问题

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一完全的整数线性规划问题。连续Hopfietd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。

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

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.