Font Size: a A A

Support Vector Regression Machine Theory And Its Industrial Application

Posted on:2007-04-14Degree:DoctorType:Dissertation
Country:ChinaCandidate:W S AnFull Text:PDF
GTID:1118360212960396Subject:Control theory and control engineering
Abstract/Summary:PDF Full Text Request
The data-based machine learning plays an important role in modern intelligent information processing area, and its main research is to draw a principle that we can not grasp at present through theoretic analysis from the observed data, then analysis, understand objective targets and predict the future data and unobservable data with this resulting principle.Support vector machine (SVM) is a new machine learning technique developed on statistical learning theory. It was firstly proposed by Vapnik and his collaborators in 1990s, and has been made progress in theoretic research, algorithm realization and applications in recent years.However, SVM, as a relatively young technique, is still imperfect. So it is vital to further develop and consummate its theory, method and extend the applications. Aiming at studying and exploring the theory and application of support vector regression machine (SVR), this thesis mainly includes the following contributions:1) An information geometry based method of SVR kernel function construction is proposed. The performance of SVM largely depends on the choice of kernel function; however, there is no theoretical guidance for the choice of kernel's type currently. Based on the analysis of geometrical structure of kernel from information geometrical viewpoint, an algorithm of kernel function construction is presented, which can make good use of the information of samples and thereby improve the performance of SVR. The application to molten steel prediction shows the effectiveness of the proposed method.2) The equivalence between SILF-SVR and ordinary Kriging is proved under the kernel framework. From the perspective of numerical analysis, SVR can be viewed as an interpolation method. After introducing the "uniform" formulations of SILF-SVR, which is obtained by using a new "uniform" loss function (soft insensitive loss function, SILF), we then testify the equivalence between SILF-SVR and another...
Keywords/Search Tags:Machine Learning, Statistics Learning Theory, Support Vector Machine, Support Vector Regression Machine, Kernel, Information Geometry, Ordinary Kriging, Domain Knowledge, Uncertainty Information, Metallurgical Industry
PDF Full Text Request
Related items