Regression Analysis And Application Of Support Vector Regression In Material Experimental Data

Statistical Learning Theory (SLT) proposed by Vapnik and co-workers from the AT & T Bell is a statistics theory for the analysis of a small-sample database. Based on STL and structural risk minimization,support vector machine (SVM) is a supervised machine learning approach and was recognized as a statistical learning apotheosis for small-sample database. SVM has shown its excellent learning and generalization ability. It has been successfully applied to solve classification and regression problems in many fields,and deliver state-of-the-art performance in real world applications.In this thesis,different kinds of regression techniques were utilized to model and predict for six material experimental datasets including column efficiency of two-dimensional chromatography,thermal expansion coefficient and oxide composition of R2O-MO-Al2O3-SiO2 system glass , 3C steel corrosion rates , wood thermal conductivity,bending strength of AlON-TiN composites,and mechanical properties of 7005 Al alloys. The prediction results and performances of support vector regresion (SVR) were compared with those of other regression methods. At the same time,the optimal synthesis parameters of AlON-TiN composites were searched out by particle swarm optimization after the optimal model between its synthesis parameters and bending strength has been established via SVR.The outline of this thesis is shown as below:①Three search methods,including grid search algorithm (GSA),genetic algorithm (GA) and particle swarm algorithm (PSO),of parameter optimization for artifical intelligence were reviewed. The advantages and disadvantages of these algorithms were introduced.②The regression principles of popular regression methods were reviewed briefly,such as least square regression (LSR),stepwise regression (SR) and partial least square regression (PLSR) among multivariate linear regression,K-nearest neighbor regression (KNNR),kernel K-nearest neighbor regression (kernel-KNNR),general regression neural network (GRNN) and artificial neural networks (ANN). The principle,algorithm,implementation,development of SVR and its applications were described in detail.③Based on the experimental datasets of column efficiency of two-dimensional chromatography , thermal expansion coefficient and oxide composition of R2O-MO-Al2O3-SiO2 system glass,3C steel corrosion rates,wood thermal conductivity, and bending strength of AlON-TiN composites,several regression methods were employed to modeling and predicting for these datasets. The results reveal that prediction performance of SVR surpasses those of other regression approaches.④The optimal process parameters of AlON-TiN composite were searched out by PSO based on the established SVR model between its synthsis process parameters and bending strength,and then the influence of multifactors on the bending strength were further analyzed.The studies of above demonstrated that,the prediction precison of SVR was superior to those of other regression methods including multivarivate linear regression,partial least square regression and neural network,and its generalization ability surpasses those of them. The results suggest that SVR is an effective and powerful technique; it may be further developed to be a potential application tool in the field of materials,such as material computer assistant design and material process parameter optimization,etc.