一种改进BP网络用于电磁兼容预测  

An Improved BP Neural Network Used for Prediction of EMC

作　　者：陈书文[1] 张煜东[2] 张斌[3] 王水花[2] 

机构地区：[1]江苏省辐射环境保护咨询中心,南京210096 [2]东南大学信息科学与工程学院,南京210096 [3]江苏省辐射监测站,南京210096

出　　处：《科学技术与工程》2009年第19期5672-5675,共4页Science Technology and Engineering

基　　金：国家自然科学基金(60872075)资助

摘　　要：为了更好地对电磁兼容进行预测,提出采用人工神经网络的方法。为了改善BP神经网络的性能,提出如下两步改进:采用剪枝法计算最佳隐层神经元数目,同时采用共轭梯度-LM算法计算网络权值。以平行线间电磁耦合干扰为具体算例,证明本文算法的预测结果的均方误差仅有10-11数量级。说明,本文算法有效。In order to predict the electromagnetic compatibility more effectively, an improved method based on artificial neural network was proposed. Two improvements were advanced for the goal of improving the performance of neural network ： on one hand the pruning method was used to get the optimal number of neurons in hidden layer, on the other hand a novel training method based on the combination of conjugate gradient and Levenberg-Marquardt was presented to calculate the weights of neural networks. The specific example on electromagnetic coupling inter- ference between two parallel wires demonstrates the median square error of the prediction is more or less only 10-11 order of magnitude. Thus, this proposed algorithm is effective.

关 键 词：电磁兼容 预测 神经网络 

分 类 号：TN911.7[电子电信—通信与信息系统]