基于拓扑结构的碱金属化合物摩尔磁化率的支持向量回归研究  

Topological research on the molar magnetic susceptibility of alkali metal compounds with support vector regression

作　　者：蔡从中[1] 庄魏萍[1] 温玉锋[1] 朱星键[1] 裴军芳[1] 肖婷婷[1] 

机构地区：[1]重庆大学应用物理系,重庆400044

出　　处：《物理学报》2009年第F06期272-277,共6页Acta Physica Sinica

基　　金：教育部新世纪优秀人才支持计划(批准号:NCET-07-0903);教育部留学回国人员科研启动基金(批准号:2008101-1);重庆市自然科学基金(批准号:CSTC2006BB5240);国家大学生创新性实验计划(批准号:CQUCX-G-2007-016)资助的课题~~

摘　　要：基于经典电动力学导出的表征简单离子磁化率的磁性点价gi所构建的分子磁性连接性指数mF及45种碱金属化合物的摩尔磁化率χm的实测数据集,利用粒子群寻优的支持向量回归(SVR)方法,建立了基于0F和1F的碱金属化合物χm的预测模型,并与基于多元线性回归(MLR)模型的计算结果进行了比较.结果显示,基于9次交叉验证的SVR模型预测的平均绝对误差、平均相对误差绝对值以及均方根误差均比MLR模型小,表明SVR模型的回归预测能力优于MLR.研究表明,磁性连接性指数mF是一种合适的分子描述符,SVR是一种预测碱金属化合物χm的有效方法.According to the experimental dataset on the molar magnetic susceptibility Xm of 45 alkali metal compounds and the topological descriptor magnetic connectivity index mF, which is extracted by the magnetic valence gi of simple ion deduced from classical electrodynamics, support vector regression （SVR） combined with particle swarm optimization for its parameter optimization is proposed to establish a model for predicting the molar magnetic susceptibility of alkali metal compound via 0F and 1F. The performance of SVR model is compared with that of multivariate linear regression （MLR） model. The results show that the mean absolute error, the mean absolute percentage error and the root mean square error for 9-fold cross validation test of SVR models are all smaller than those achieved by MLR models. It is revealed that the generalization ability of SVR model is superior to that of MLR model. This study suggests that magnetic connectivity index is an effective descriptor and the SVR is a powerful approach to the prediction of the molar magnetic susceptibility of alkali metal compounds.

关 键 词：碱金属化合物 摩尔磁化率 支持向量回归 预测 

分 类 号：O482.5[理学—固体物理]