材料期刊网

将偏最小二乘法与BP人工神经网络结合,建立了一种新的预测模型：PLS-BP神经网络模型。模型应用偏最小二乘法来提取主成分R及得分T,根据交叉有效性检验和留N法来确定PLS的成分个数,PLS-BP网络的输入数目和网络隐含层的节点数目,最终确定网络的结构为6-11-1。应用该模型可以有效地避免几个因素之间的多重相关性问题,同时也能更好地解决非线性问题,克服了偏最小二乘和单纯BP网络的缺点。在钢筋屈服强度的预测中表明,应用PLS-BP模型预测的误差均小于1.03%,比应用于偏最小二乘回归模型的误差6.19%要小得多,并且预测值和实际值比较吻合。

Combining with the partial least squares and artificial network , a new prediction model is established-- PLS-BP neural network model. Partial least squares regression is applied to extract principal components R and score T, according to the cross validation and left N method, PLS component number, input of the PLS-BP neu- ral network and the number of hidden layer nodes of network is determined, ultimately is the network structure for 6-11-1. Using the model, factors between multiple correlations can be effectively avoided; at the same time, the non-linear problem can be solved better and partial least squares and simple BP network faults can be overcome. The yield strength of steel projections shows that the error predicted by application of PLS-BP model is less than 1.03% ; it is much smaller than the partial least squares regression error of 6.19% and coincides with the actual value.