Reduced Least Squares Support Vector Based on Kernel Partial Least Squares and Its Application Research

Abstract

Firstly, a rapidly reducing kernel matrix method to construct sparse least squares support vector machines is proposed. By minimizing the Euclidean distance between the mapping of sample vector in original feature space and linear combination of base in reduced feature space, the columns in kernel matrix are eliminated according to a order array which are composed of the maximum of every column in original kernel matrix, so that the reduced kernel matrix is sparse. Then, the parameters of reduced least squares vector machine are identified by kernel partial least squares. Lastly, a nonlinear dynamic prediction model using reduced least squares support vector machine on the base of kernel partial least squares is constructed to predict the total converting time of copper converter blowing time during slag making period. The simulation results show that the reduced least squares support vector machine based on kernel partial least squares has the performances like, better efficiency of computation, accuracy of prediction and preferable application value.