Research And Application Of Hybrid Time Series Model Based On Support Vector Machine

Stock price index is a complex system of nonlinear dynamics. Stock price movements to predict not only beneficial to a country of their own to implement favorable macroeconomic control policies, but also to provide a reference for equity investors, to reduce investment risk. General stock prediction method existed learning dimension disaster, easy to fall into local optimum. Support Vector Machine(SVM) can better resolve these issues but because of their own model to calculate complicated process so its use is limited by long-term, based on this, the support vector machine model to simplify the use of phase space reconstruction process of the calculation, and then the non-linear optimization problem is transformed into linear optimization problems, least squares support vector machine(LSSVM) based on this production.Present least squares support vector machines model focused on two aspects, the sample sparsity and inappropriate parameter selection problem, all cause the machine to poor generalization performance, low prediction accuracy. First, for the sparsity problem that this article algorithm using simple, less parameters, capable of handling large data sample of K-mean clustering algorithm to sample data excluding the classic strong correlation redundant data, the characterization of point set represents the sample space, better to improve the sparsity of the sample. Secondly, the choice of particle swarm(PSO) algorithm to find the optimal parameter value LSSVM model, select the maximum fitness value parameter value, then the generalization ability of the model is the strongest. Improved PSO optimization based on final proposed LSSVM stock price prediction model, and select Dawn shares the latest stock data May 23, 2014 to April 30, 2015, the use of MATLAB software simulation to verify the validity of the established model. The results showed that: stock prices forecasting model based on improved PSO LSSSVM LSSVM model compared to less support to calculate speed, high prediction accuracy.Based on the theory of support vector machine, and combined with the general model of time series, this paper establishes a hybrid time series model based on support vector machine in 2014 04, 03, 26, 15,, as the experimental data, and compared with the real value, and with a single time series model and support vector machine model prediction results.