论文部分内容阅读
以86个离子化合物的正、负离子的有效核电荷Z~(*+)、Z~(*-)、离子半径r_+、r_-,以及正离子的荷径比Z~(*+)/Υ_+5种结构参数作为自变量,以晶格能U作为因变量,采用BP神经网络建立关于无机离子晶体晶格能的结构-性质关系(QSPR)模型.该模型由输入层、隐含层和输出层构成3层BP神经网络,86个离子化合物样本则按文献分别划分为训练集和验证集.研究表明,当隐含层神经元个数为5时模型效果最佳:该模型对训练集拟合结果的决定系数R~2=0.9965,平均相对误差MRE=1.63%;对验证集预测结果的R~2=0.9952,MRE=1.85%.
The effective nuclear charges Z ~ (* +), Z ~ (* -), ionic radius r_ +, r_- of positive ions and negative ions of 86 ionic compounds and the positive charges Z ~ (* +) / Y_ +5 kinds of structural parameters as independent variables and lattice energy U as the dependent variable, the structure-property relationship (QSPR) model of lattice energy of inorganic ionic crystals was established by using BP neural network.The model consists of input layer, hidden layer and The output layer constitutes a 3-layer BP neural network, and 86 ion-compound samples are divided into training set and verification set respectively according to the literature.The research shows that the best model is when the number of hidden layer neurons is 5: The determination coefficient of the fitting result is R ~ 2 = 0.9965, the average relative error is MRE = 1.63%; R ~ 2 = 0.9952 and MRE = 1.85% for the validation set.