小波范数熵特征提取的模拟电路故障诊断方法  

A Fault Diagnosis Approach of Electric Circuit Based on Wavelet Norm Entropy as Feature Extractor

作　　者：肖迎群[1] 何怡刚[1,2] 张广辉[1] 

机构地区：[1]贵州理工学院电气工程学院,贵州贵阳550003 [2]合肥工业大学电气与自动化工程学院,安徽合肥230009

出　　处：《机械与电子》2015年第6期3-9,共7页Machinery & Electronics

基　　金：贵州省科学基金重点项目(黔科合LH字[2014]7356号);贵州理工学院高层次人才科研启动经费项目(XJGC20131203)

摘　　要：提出一种遗传优化神经网络与小波范数熵相结合的新型模拟电路故障诊断方法,降低神经网络的结构冗余度和减少过拟合现象。小波范数熵方法提取了故障数据的本质特征,遗传算法优化了神经网络的体系结构,诊断系统实施了模拟数据的故障分类。仿真结果表明,同小波变换预处理的故障诊断系统相比较,这种诊断系统具有更好的网络收敛性能、更高的诊断精确度和更强的推广能力,能对模拟电路故障元件进行有效识别和分类。A fault diagnosis method combing wave- let norm entropy theory and genetic neural networks is introduced in order to reduce structural redundancy of neural networks and its over - fitting of nonlinear ap- proximation. Wavelet norm entropy theory is used to extract the intrinsic features of fault data, and a genet- ic algorithm is utilized to reduce the architecture of the neural network and a fault diagnosis system of analog circuit is constructed to execute fault classification. Simulation results show that in comparison to wavelet transformation pretreatment the proposed system has better convergence performance, higher diagnosis ac- curacy and better generalization ability. Ultimately,the system can effectively recognize and classify the faulty components of electric circuits.

关 键 词：小波理论 范数熵 遗传算法 神经网络 故障诊断 

分 类 号：TN707[电子电信—电路与系统] TP183[自动化与计算机技术—控制理论与控制工程]