(中国科学院上海冶金研究所，上海 200050)
摘 要: 应用原子参数-人工神经网络研究了MM'O3型的复氧化物的熔点与原子参数之间的关系，并利用已知样本集训练的人工神经网络对MM'O3型复氧化物的熔点用“留一法”进行了预报，预报结果与实测值符合较好，误差一般小于5%。研究结果表明，选择适当的原子参数-人工神经网络算法可以用于M2O3-M'2O3系形成的MM'O3型复氧化物的熔点的预报。
关键字: 氧化物系 相图计算 复氧化物
（Shanghai Institute of Metallurgy, Chinese Academy of Sciences, Shanghai200050, P. R. China）
Abstract:ANN-atomic parameter method was used in investigating the regularity of the melting point of MM'O3 type intermediate compounds. A method for the prediction of the melting points of MM'O3 type intermediate compounds was proposed. And three atomic parameters of constituent elements, ionic radii Ri, electronegativity X, covalent radii Rc and the known melting points data of MM'O3-type compounds were used to find a mathematical model for the melting point prediction, with ANN as the method of computation. The predicted results agree well with the experimental ones.
Key words: oxide system calculation of phase diagram complex oxide
