(Zr0.7Sn0.3)TiO4陶瓷性能预报的支持向量回归模型

将支持向量回归方法用于(Zr0.7Sn0.3)TiO4陶瓷的配方性能关系研究中,分别建立了(Zr0.7Sn0.3)TiO4陶瓷介电常数和损耗角正切的支持向量回归模型,并与逆传播人工神经网络、多元线性回归模型进行了比较.用留一法分别检验了支持向量回归、逆传播人工神经网络和多元线性回归3种不同模型的预报能力,结果表明:上述3种模型对于(Zr0.7Sn0.3)TiO4陶瓷介电常数的留一法预报的平均相对误差分别为1.083%、1.632%、1.931%,对于损耗角正切的留一法预报的平均相对误差分别为0.999%、1.746%、1.414%.因此,支持向量回归模型的预报能力较好,可望在陶瓷配方设计中的多变量、非线性问题和小样本体系中发挥较好的作用,为新型介电陶瓷的性能预报和配方优化提供一条全新可靠的途径.

Prediction of Properties of (Zr0.7Sn0.3)TiO4 Dielectric Ceramics Using Support Vector Regression

The support vectorregression(SVR) is appliedtoinvestigatethe relationship betweenthe propertiesandformulation of(Zr0.7Sn0.3)TiO4dielectric ceramic.The models for predicting dielectric constant and lossangletangent are proposed respectively.Using the leave-one-out cross-validation method,the mean relativeerrors(MRE) of dielectric constant in SVR,back-propagation artificial neural network(BP-ANN) andmultiple linear regression(MLR) models are checked,withthe results of 1.083 %,1.632 %and 1.931 %respectively.Similarly,the MRE of loss angle tangent in SVR,BP-ANNand MLRare 0.999 %,1.746 %and 1.414 %respectively.It istherefore demonstratedthat support vector regressionis a useful tool in dealingwith multi-variable problems with non-linearity or small size of data set inthe formulation design of dielectricceramics.