摘要
由于ＢＰ网络用于材料研究时存在“过拟合”现象，本文提出一种以拟合误差曲线和预报误差曲线为依据的选择网络参数，优化网络结构的方法。在一定程度上克服了ＢＰ网络的“过拟合”缺点，并将此方法应用于ＺｎＯ压敏陶瓷材料的性能预报，效果良好。
A new method based on fitting error curve and forecast error curve applied to the selection of BPN(backpropagation network) parameters and optimization of BPN structure is proposed to overcome the shortcoming of overfitting in BPN to a certain extent. The optimal BPN is used to forecast the electric properties of ZnO varistor ceramics. Satisfactory results show that BPN can be applied to the fileds of electronic ceramics research.
