不同神经网络在橡胶配方性能预测中的应用研究

Experimental Study on Different Neural Networks in the Prediction of EPDM Formulation Performance

选取EPDM配方中的物料的用量作为输入,以硫化胶的基本性能作为输出,通过正交实验取得的16组实验数据作为样本数据,并以其中1~13组数据为训练集,14~16组数据为测试集,建立四种人工神经网络预测模型。通过对比不同模型预测的均方误差、最大误差、最小误差和平均误差来判断神经网络的表现。结果表明:BP神经网络表现最好,对拉伸强度、拉断伸长率和撕裂强度的预测精度都较高,其次是ELMAN和RBF神经网络,对拉断伸长率和撕裂强度有很高的预测精度,而GRNN神经网络容易发生过拟合,不适用于胶料配方性能的预测。

The performance of four kinds of neural networks in predicting the performance of EPDM were com- pared. The experimental data were obtained by orthogonal experiment. The neural network was trained by 13 groups, and the other three groups were tested on the neural network to compare the mean square error, the maxi- mum error, the minimum error and the average error of the neural network prediction results. The results showed that the BP neural network was the best, and the prediction accuracy of tensile strength, elongation at break and tear strength were high,followed by ELMAN and RBF neural network, and the elongation and tear strength were high pre- diction accuracy,while the GRNN neural network was not suitable for the performance prediction of the compound.