Research And Application Of The Temporal Data Forecasting Based On LS-SVM

In the real world, information systems often involve the temporal data, which is correlated with time and contains the time attribute of the data. Temporal data is according to a certain time differentiates afresh with time data, used to study the potential rule characteristic data. The temporal data through certain way after transmutation, in which people often can find the rules that the original time series does not have. With the increasingly large amount of temporal data, as well as the needs to find the potential rules and information hidding in the data, temporal data mining has been studied.Support vector machine (SVM) is a new technology based on the statistical learning theory and principle of structural risk minimization, to solve the nonlinear and uncertain problems in the temporal data mining. Least squares support vector machine (LS-SVM) is a kind of improved support vector machine. Through a series of transmutation, least squares support vector machine eventually transform the quadratic programming problem of the standard SVM into solving linear equations, which improve the precision and the speed of convergence.Based on the depth analysis of the concept of temporal data, temporal type, temporal granularity as well as the support vector machine, used by the method that the atomic temporal type constructs other temporal type,we wil construct a p-temporal granularity of temporal data model, and set the initial training parameters of least squares support vector machine--adjustable parameter r controling regression error and RBF kernel function parameter s2, then standardize the primitive stock data of financial market. Because the parameters r & s2 of least squares support vector machines must be given before the training by the user, and there is no certain theoretical basis, with certain subjectivity and a priori, therefore, in order to get a better forecasting effect, this paper improves an optimization algorithm based on swarm intelligence -- particle swarm optimization (PSO) algorithm, optimizes on the the model parameters. Then,we will train the model of least square support vector machine, finally give the temporal data forecasting based on least squares support vector machines.The temporal data forecasting based on least squares support vector machine can not only solve the nonlinear and uncertain problems in the temporal data mining, but also improve the solution accuracy and the speed of convergence. We will apply the modal to the financial industry, analysis the financial market temporal data to find the valuable information and rules, which has important application values.