偏最小二乘法-人工神经网络用于复氧化物形成条件的判别

DISCRIMINATION OF THE FORMATION CONDITIONS OF COMPLEX OXIDE USING PARTIAL LEAST SQUARES METHOD AND ARTIFICIAL NEURAL NETWORK METHOD

本文应用人工神经网络算法结合偏最小二乘算法,对两元氧化物系化学键参数进行模式识别分析,用计算机对两不同的氧化物间是否有复合氧化物生成进行分类和预报,得到满意的结果。为了确定两氧化物间是否有复氧化物生成,本文首先用PLS回归算法初选化学键参数,通过样本点在PLS正交分解矢量平面上的投影图,按形成与不形成对两元氧化物进行分类,从而确定一组对复氧化物形成影响最大的化学键参数,实现对能否生成复氧化物进行判别。最后,用选出的化学健参数训练人工神经网络,训练好的网络就可以对复氧化物的形成与否进行预报。

Chemical bond parameters of binary oxide system are recognised by use of artificial neural network method combined with partial least squares method. Classification and prediction of complex oxides formed in two different oxides are carried out by use of computer with satisfactory results. In order to find out if there exist complex oxides between two oxides, the following steps are used: 1. A group of optimal bond parameters are selected using the PLS algorithm. 2. The selected bond parameters are used to train the backpropagation artificial neural networks. 3. The trained B-P neural networks are used to predict the occurrence of complex oxide.