Standard Entropies Of Cations In Solid Compounds With Support Vector Regression

Support vector machine (SVM), proposed by Vapnik and co-workers in 1995, is based on structural risk minimization principle and statistical learning theory. It has been well-known as a paragon to learn in case of a small number of samples, and has a strong learning and generalization ability. SVM has been successfully applied to solve classification and regression problems in many fields.The regression prediction results of standard entropies of 70 kinds of solid compounds'cations by using different descriptors and different regression methods (linear/nonlinear approaches) have been summarized. Furthermore, support vector regression (SVR) is proposed to predict the standard entropies of 70 kinds of solid compounds'cations. And then the prediction results of SVR are compared with those achieved by using other regression methods, including multi-variable linear regression (MLR) and artificial neural networks. Meanwhile, SVR is proposed to predict the density of selective laser sintering parts sintered under different sintering conditions, and the predicted results of SVR is also compared with that of BPNN.The main contents of this thesis are as follows:1) The principle of machine learning methods and the basic idea of statistical learning theory are introduced briefly, some core contents of statistical learning theory are addresses simplicity, and then the principle of support vector regression (SVR) and the optimization method of its parameter are introduced. The algorithms of fuzzy-SVR and weighted-SVR have also been introduced briefly.2) The two common methods for parameter optimization (simulated annealing algorithm and ant colony algorithm), and several other common regression approaches, such as probabilistic neural network (PNN), MLR and ridge regression are introduced. Their merits and weaknesses are discussed briefly.3) The basic concept of entropy and its fundamental nature are in a brief introduction. The physical meaning of entropy on the thermodynamic entropy, principle of equivalent on statistical entropy, the thermodynamic principle of entropy increase of entropy, statistical entropy, as well as the Boltzmann entropy and Clausius entropy, all are summarized briefly.4) SVR is applied to model & predict action standard entropies of 70 solid compounds and the density of selective laser sintering parts according to the related experimental dataset, and their prediction results have been analyzed and compared with each other.The prediction accuracy by SVR is superior to those predicted by other regression approaches, such as MLR and ANNs, and the generalization ability of SVR surpasses those of other methods, which has been demonstrated by the studied results. It is revealed from this study that SVR is an effective tool in processing of experimental data. It is expected that a better progress on the development and application of SVR would be taken in other physical experimental data.