支持向量机算法在熔盐相图数据库智能化中的若干应用  

Support vector machine applied to intelligent database of phase diagrams of molten salt systems

作　　者：包新华[1] 陆文聪[1] 陈念贻[1] 

机构地区：[1]上海大学理学院化学系计算机化学研究室,上海200436

出　　处：《计算机与应用化学》2002年第6期723-725,共3页Computers and Applied Chemistry

基　　金：国家自然科学基金委和美国福特公司联合资助(9716214)

摘　　要：支持向量机(SVM)方法可被用于熔盐系未知相图的计算机预报。将已知的二元卤化物系相图数据作为训练集,体系组分的离子半径和电负性作为特征量,用SVM方法可预报中间化合物的形成与否、熔化类型(同分熔化还是异分熔化)和估计中间化合物的熔点。本文报道了M_2M'F_4型的中间化合物的形成判据、M_3M'Cl_6型化合物的熔化类型的判据以及MM’X_4型中间化合物熔点计算的回归方程式。用“留一法”检验所得的数学模型并将结果与传统的模式识别方法(Fisher法和KNN)进行了比较,结果表明:SVM的预报准确率比Fisher法和KNN法都高。因此,SVM方法有望成为计算机预报未知相图的有力工具。In this work, SVM (support vector machine) method was used for the computerized prediction of unknown phase diagrams of molten salt systems. Using the data of known phase diagrams of binary halide systems as training set, and some atomic parameters related to ionic radi-i, electronegativity as features, SVM method was applied to predict the formability, melting type and melting points of intermediate compounds. The criteria for prediction such as formation of intermediate compound of M2M'F4 type, the melting type of M3M'Cl6 type compounds and the melting points of MM'X4 type compounds were obtained. And the results of the cross-validation experiment comparing SVM with two other pat-tern recognition methods, Fisher method and KNN method, indicated that the predictive ability of SVM was better than that of Fisher method or KNN method.

关 键 词：支持向量机算法 熔盐 相图 数据库 智能化 计算机预报 中间化合物 原子参数 

分 类 号：O642.42[理学—物理化学] O6-39[理学—化学]