人工神经网络法对MeX－REX_3熔盐系相图特征的预报

ARTIFICIAL NEURAL NETWORK APPLIED TO COMPUTERIZED PREDICTION OF MeX-REX_3 TYPE PHAES DIAGRAMS

用化学键参数－人工神经网络方法对ＭｅＸ－ＲＥＸ。熔盐系相图进行研究，总结相图若干特征的规律．结果表明，人工神经网络可判断中间化合物的形成，化学配比以及熔化类型，并能预报中间化合物的熔点，分解温度和包晶点的组成，结果与实测相图符合较好．其研究结果已纳人材料设计专家系统．

Artificial neural network with chemical bond parameters as inputs is applied to MeX-REX3(Me is monovalent metal,RE rare earth metal, X halides)type phase diagrams investigation. After training with the data of known phase diagrams, the mathematical models obtained can be used to make computerized prediction of the formability,the chemical stiochiometry,the meltting types(congruent or incongruent melting),the melting points or decomposition temperatures of intermediate compounds and the compositions at the peritectic points of peritectic reaction in unknown molten salt phase diagrams,with good results.