模拟电路故障诊断的BP神经网络方法  

A Method for Fault Diagnosis of Analog Circuit by BP Neural Network

作　　者：何怡刚[1] 梁戈超[1] 

机构地区：[1]湖南大学电气与信息工程学院,湖南长沙410082

出　　处：《湖南大学学报（自然科学版）》2003年第5期35-39,共5页Journal of Hunan University:Natural Sciences

基　　金：高等院校博士学科点专项科研基金[20020532016];湖南大学撷英计划资助.

摘　　要：提出了一种采用改进型的BP神经网络诊断模拟电路故障的新方法.介绍了包括附加动量法和自适应学习速率法在内的改进型的BP网络算法,其中附加动量法是在每一个权值的变化上加上一项正比于前次权值变化量的值,不仅考虑了误差在梯度上的作用,而且考虑在曲面上变化趋势的影响;自适应学习速率是根据误差函数值的变化对学习率进行实时调整,可以保证网络总是以最大的学习速率学习 最后,本文给出了仿真实例Some methods are put forward to improve the Algorithm of BP Neural Network, and are applied to the fault diagnosis in circuit. This method includes the additional momentum algorithm and selfadaptive study speed algorithm. The additional momentum algorithm is adding an item which is proportional to the value of fore weights variation. The algorithm considers not only the impact of error in the gradient, but also the impact of change trend of error. Selfadaptive study speed algorithm is realtime adjustment of study speed according to the variation of the error function value, which ensures that the network studies at the maximal study speed. Finally, simulation results show the improved Algorithm of BP are more feasible and efficient in contrast with the traditional methods.

关 键 词：故障诊断 神经网络 BP算法 

分 类 号：TN46[电子电信—微电子学与固体电子学]