基于Bayesian正规化BP神经网络的粘结NdFeB永磁体性能预测  

Property Prediction of Bonded NdFeB Permanent Magnet on Bayesian-regularization BP Neural Network

作　　者：储林华[1] 查五生[1] 刘锦云[1] 刘桂明[1] 周晓庆[1] 张静怡[1] 

机构地区：[1]西华大学材料科学与工程学院,四川成都610039

出　　处：《稀土》2009年第2期39-42,50,共5页Chinese Rare Earths

基　　金：四川省教育厅重点项目(2004A110)

摘　　要：基于MATLAB平台和现有的少量实验数据,采用Bayesian正规化法,建立了一个输入为工艺参数、输出为NdFeB永磁体性能参量的BP(Back Propagation)神经网络预测模型,并通过测试样本检验了ANN(Artificial NeuralNetwork)模型的准确性。实验表明,建立的Bayesian正规化BP神经网络模型不仅能准确地拟合训练值,而且能很好地预测未知样本,将该模型应用于材料制备工艺设计,可以明显缩短实验周期,提高工艺设计效率,对实际的研究工作具有一定的指导意义和应用价值。A BP neural network model was built for the bonded NdFeB magnet using the Bayesian regularization via Matlab and limited data. The fabricating parameters were taken as input data with magnetic properties as the output data. The accuracy of ANN model was evaluated by the sample test. The experimental results show that the model can not only exactly imitate training valuation but also make prediction accurately. And making use of the model in material preparation crafts optimizing, we can obviously shorten the experimental period and improve the craft designing effeciency. Therefore, the modeling method is effective and the model is available.

关 键 词：Bayesian正规化 BP神经网络 NDFEB粘结磁体 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]