Parametric Model Based on GA and SVM

A new method to develop a parametric model based on genetic algorithm (GA) and support vector machines (SVM) is proposed. The proposed method is achieved in three steps. In the first step, the non-stationarity of the series is identified. If the time series is stationary, the second step is executed directly. If the time series has the characteristics of non-stationarity, the non-stationary time series is processed to become a stationary time series by trend extraction technique and then the second step is executed. In the second step, GA is used to determine the primary order of the parametric model. In the last step, the order of the parametric model is determined further using SVM on the basis of the result of the second step and hence the final parametric model is developed. GA is adopted to construct the rough frame of the parametric model, which reduces the task of SVM. SVM is produced to improve the generalization performance of the parametric model obtained based on GA in the second step. The simulation result shows that the proposed method outperforms the single GA and single SVM.