Chaotic wind speed series forecasting based on wavelet packet decomposition and support vector regression

According to the chaotic and non-stationary characteristics of wind speed, this paper proposes an approach to forecast the wind speed for 1h-6h ahead at a resolution of every 10min based on wavelet packet decomposition, phase space reconstruction and support vector regression (SVR). The mentioned three methods play different roles in the prediction process respectively. Firstly, the method of wavelet packet decomposition is used to decompose the original wind speed series into different frequency components. Then different SVR models are built up based on phase space reconstruction and the subsequences are predicted with the method of SVR respectively. At last, the final predicted wind speed can be calculated by the superposition of respective predictions. Two test data sets with different features are presented to illustrate the effectiveness. The results indicate that the forecast accuracy of the proposed approach is greatly improved compared with other presented approaches, and there is a big promotion and application value in practice.