A New Intelligent Model for Nonlinear Time Series Prediction

Abstract

The problem of nonlinear time series prediction using integrated intelligent methods based on support vector machine (SVM) and particle swarm optimization (PSO) is studied. Aiming to the open problems of nonlinear time series prediction such as the best number of historical points and parameters of SVM are hard to be determined, a novel model for time series prediction based on PSO and SVM models is proposed. For the task of improving precision of time series prediction, the basic idea is to construct model integrated both the advantages of PSO with powerful intelligent global optimization capability and SVM with excellent prediction capability. The model is a self-adaptive parameters optimizing one through using PSO algorithm to search the global optimum values of number of historical points and parameters of SVM during the training process of SVM. Experiments results of two benchmark data sets including Mackey-Class time series data and Santa Fe chaotic laser data prove the feasibility and good effectiveness of the model for nonlinear time series prediction.