基于l_2正则化回声状态网络的模拟电路故障诊断  

Analog Circuit Fault Diagnosis Based on l_2 Regularization Echo State Network

作　　者：王洪[1] 牛晓灵 WANG Hong ,NIU Xiaoling(Department of Computer, Pingdingshan Industrial College of Technology, Pingdingshan Hen＇an 467001, Chin)

机构地区：[1]平顶山工业职业技术学院计算机系,河南平顶山467001

出　　处：《电子器件》2017年第5期1283-1286,共4页Chinese Journal of Electron Devices

基　　金：河南省科技厅重点科技攻关项目(152102310463)

摘　　要：为了提高模拟电路故障诊断的准确率,提出了一种基于l_2正则化回声状态网络(l_2-RESN)的模拟电路故障诊断方法。l_2-RESN在ESN的约束优化函数中引入l_2正则化因子,推导带正则化因子的ESN输出权重计算公式,避免传统的ESN算法因矩阵奇异而降低模型泛化能力。实验结果表明,相比于支持向量机(SVM)和标准ESN,l_2-RESN的诊断准确率分别提高1.11%和18.34%。证明l_2-RESN能够有效提高模拟电路诊断的准确性。In order to improve analog circuit fault diagnosis accuracy,an analog circuit fault diagnosis algorithm based on l_2 regularization echo state network（ l_2-RESN） is proposed. l_2 regularization factor is introduced into ESN constrained optimization function. ESN output weight computation formula with regularization factor is deduced,which avoids the problem of reducing the model generalization ability due to singular matrix. Experiment result shows that,comparing with support vector machine（ SVM） and standard ESN,l_2-RESN fault identification accuracy rate increases 1.11% and 18. 34% separately. The result indicates that l_2-RESN improves analog circuit diagnosis accuracy effectively.

关 键 词：模拟电路 故障诊断 神经网络 拉格朗日函数 

分 类 号：TN710[电子电信—电路与系统]