Wind power prediction is one of the most critical aspects in wind power integration and operation. This paper presents a new approach to a wind power prediction by combining support vector regression (SVR) with a local prediction framework which employs the correlation dimension and mutual information methods used in time-series analysis for data preprocessing. Local prediction makes use of similar historical data patterns in the reconstructed space to train the regression algorithm. To build an effective local SVR method, the parameters of SVR must be selected carefully. Therefore, a new method is proposed in this paper. The proposed method which known as genetic algorithm (GA)-Local SVR searches for SVR's optimal parameters using real value GA. These optimal parameters are then used to construct the local SVR algorithm. The performance of the proposed method (GA-Local SVR) is evaluated with the real world wind power data from England and is compared with the seasonal auto regressive integrated moving average (SARIMA) method and radial basis function (RBF) network. The results show that the proposed method provides a much better prediction performance in comparison with other methods employing the same data.