基于LM-BP神经网络算法的模拟电路故障诊断

Analog Circuits Fault Diagnosis Based on LM-BP Optimized Neural Network

针对BP神经网络在模拟电路故障诊断中存在的网络学习收敛速度慢、不易获得全局最优解、诊断精度低以及网络结构不确定等缺点,采用遗传算法对BP神经网络结构、初始连接权值和阈值进行全局优选,并利用Levenberg-Marquardt算法训练BP网络以克服这些缺陷;选取ITC97中的Elliptical Filter电路作为测试电路,设置故障并进行诊断,仿真实验表明,该诊断方法能够有效诊断模拟电路中存在的故障,并且具有更高的诊断精度。

There are inherent disadvantages in traditional BP neural network.First,the error of training drops slowly.Second,the adjustment time is too long because of too many iteration steps.Last but not least it even easily falls into local minimum and is hardly able to extricate itself,which leads to low accuracy of diagnosis.Therefore,a new BP network method optimized by genetic algorithm（GA）and Levenberg-Marquardt（LM）algorithm is proposed.In this method,the BP network＇s structure is optimized by GA.Then LM algorithm is used to train the BP network.The training result could diagnose the faults of analog circuits,which is able to overcome the inherent disadvantages of traditional BP network.Several simulation and experimental results are presented,demonstrating the effectiveness and applicability of the developed method.