基于遗传BP网络的模拟电路故障诊断方法及其应用  

A GA-BPNNs Based Approach for Fault Diagnosis of Analog Circuits and Its Application

作　　者：祝文姬[1] 何怡刚[1] 

机构地区：[1]湖南大学电气与信息工程学院,长沙410082

出　　处：《计算机辅助设计与图形学学报》2009年第9期1283-1289,共7页Journal of Computer-Aided Design & Computer Graphics

基　　金：国家自然科学基金(50677014;60876022);高校博士点基金(20060532002);国家"八六三"高技术研究发展计划(2006AA04A104);湖南省自然科学基金(2008GK2022)

摘　　要：针对BP网络诊断模拟电路故障时存在网络结构复杂且可能出现误诊断的不足,提出一种小波变换、遗传算法与神经网络相结合的模拟电路故障诊断的新方法.该方法使用节点电压信号经小波变换、主元分析和归一化处理来实现故障特征的提取,以减少信号的冗余;由于BP网络易陷入局部最优,采用遗传算法来优化BP网络的结构和初始权值分布,使优化后的神经网络具有较好的收敛性能.最后结合电路实例,对文中提出诊断方法的原理与实现进行了较深入的分析,建立了该方法的测试系统,并通过工程应用效果进一步验证了文中方法的正确性.The application of the conventional back propagation neural networks （BPNNs） in the fault diagnosis of analog circuits might lead to a complex architecture for the NN and false diagnosis. A new analog fault diagnosis method, which is based on wavelet decomposition, genetic algorithm （GA） and neural networks （NNs）, is proposed in this paper. The proposed method uses wavelet transform,principal component analysis （PCA） and normalization to deal with the node voltages of the circuit under test （CUT） to extract fault features, where the overlap of the features can be minimized. Under considering the shortcomings of that BPNNs easily fall into local minima, the proposed approach selects GA to optimize the structure and original weight distribution of BP networks. Finally, the experiments of the applications of our proposed method are expounded in this paper. The testing system based on the proposed method is built, and the application results further verified the effectiveness of our proposed method.

关 键 词：BP网络 模拟电路 故障诊断 小波变换 遗传算法 

分 类 号：TN277[电子电信—物理电子学]