摘要
作为电子系统的重要组成部分,对模拟电路进行故障诊断研究很有必要,而模拟电路元件具有的非线性、离散型、可用于检测的节点数量少等因素,传统的故障诊断方法效果并不理想。因此,进一步探究模拟电路故障诊断方法就已然成为关注重点。神经网络技术因其特性在模拟电路的故障诊断方面表现良好,但传统的神经网络算法的局限性影响了其在实际诊断中的应用。针对反向传播(Back Propagation, BP)神经网络存在的固有缺陷:收敛速度慢、容易陷入局部极小,本研究采用Levenberg-Marquardt(LM)算法对BP神经网络进行训练,代替了传统的BP神经网络学习算法中的梯度下降法,进而寻找最优的网络连接权值。通过仿真实验表明,此方法能够改善BP神经网络的稳定性和学习效率,大幅提升了对模拟电路故障诊断的准确度,同时也有效加快了网络的收敛速度。
As an important part of electronic systems,it is necessary to study fault diagnosis of analog circuits.However,traditional fault diagnosis methods are not ideal to the nonlinearity,discreteness of analog circuit components,and the limited number of nodes available for testing.Therefore,it is particularly important to further explore fault diagnosis in analog circuits.Neural network technology has performed well in fault diagnosis of analog circuits due to its characteristics.However,the limitations of traditional neural network have affected their application in actual diagnosis.To overcome the inherent defects of Back Propagation(BP)neural networks,such as slow convergence and tendency to get trapped in local minima,study employs the Levenberg-Marquardt(LM)algorithm to train the BP neural network,replacing the traditional gradient descent method in BP neural network learning,thereby seeking the optimal network connection weights.The simulation results show that this method can effectively improve the stability and learning efficiency of BP neural network,greatly improve the accuracy of analog circuit fault diagnosis,and accelerate the convergence rate of the network.
作者
邵丽萍
杨正亿
罗双
吴丽娟
赵芳云
SHAO Liping;YANG Zhengyi;LUO Shuang;WU Lijuan;ZHAO Fangyun(School of Information Engineering,Guizhou University of Engineering Science,Bijie 551700,Guizhou,China;School of Science,Guizhou University of Engineering Science,Bijie 551700,Guizhou,China)
出处
《智能计算机与应用》
2025年第5期90-96,共7页
Intelligent Computer and Applications
基金
贵州省教育厅高等学校科学研究项目(黔教技[2022] 402号)
毕节市科学技术项目(毕科联合[2023] 50号)
贵州省教育厅2024年度自然科学研究项目(黔教技[2024] 256号)。
关键词
模拟电路
故障诊断
BP神经网络
LM算法
梯度下降法
analog circuit
fault diagnosis
BP neural networks
LM algorithm
the gradient descent method
