Study On Wavelet Support Vector Regression Model And Application

As the realization of structure risk minimization, support vector machine (SVM) has the advantages of global optimum, simple structure, and good generalization property. In this dissertation, some problems of SVM algorithms are analyzed and the reasons that caused these problems are analyzed, and two types of new algorithms of SVM are presented. Attention is paid on the problem of selection and construction of kernel functions. Based on wavelets analysis and support vector machine, four different types of SVM are proposed. The main research in this paper can be classified as follows:(1) Generally, imperceptible features are very important in non-stationary signal processing. Some kinds of methods combined wavelet analysis with kernel function technology to deal with this kind of signals are discussed in this dissertation. The feasibility of constructing kernel functions using wavelets basis is studied. The constructing method of wavelet basis kernel function is studied, and three kinds of kernel machine models are presented based on the linear programming SVM algorithms. Numerical simulation experiments are carried to validate its correctness. Simulation results show that the three wavelets kernels presented in this dissertation have very desirable performance than that of Gaussian kernel and polynomial kernel.(2) As the basis of SVM algorithms, the VC dimension theory and structure risk minimization construct the theoretic frame work for the improvement of statistics prediction method and empirical nonlinear predictive methods. Under this frame work, this dissertation rebuilds prediction method, and further improves SVM algorithms. Based on reproducing kernel theory in the Hilbert space, a new scaling reproducing kernel function is constructed in the Hilbert space. The proposed method has been adapted to the forecast of traffic flow. Experiment results show that the SVM with the proposed kernel function has better ability of generalization.(3) Asynchronous machine is a typical nonlinear plant. The parameter identification and the self-tuning of the controller parameters of the asynchronous machine are difficult problems. When online learning machine is...