Multi-step ahead wind speed forecasting using a hybrid model based on two-stage decomposition technique and AdaBoost-extreme learning machine

Accurate and reliable wind speed forecasting is of great importance to wind power generation and integration. This paper shows the development of a novel hybrid model for wind speed forecasting and demonstrates its efficiency. In the proposed hybrid model, a two-stage decomposition algorithm combining the complementary ensemble empirical mode decomposition with adaptive noise (CEEMDAN) and the variational mode decomposition (VMD) is introduced to deal with the nonlinearity of wind speed time series. The CEEMDAN is exploited to decompose the original wind speed series into a series of intrinsic mode functions (IMFs) with different frequencies. The VMD is employed to re-decompose the IMF with the highest frequency using CEEMDAN into a number of modes successively. And then an improved AdaBoost.RT algorithm is coupled with extreme learning machine (ELM) to forecast all the decomposed modes using CEEMDAN and VMD. The AdaBoost.RT technique is improved by updating the threshold value self-adaptively. Finally, the forecasting value of the original wind speed series is obtained by adding up the forecasting results of all the decomposed modes. The proposed hybrid model has been applied to four datasets of wind speed observations for one-, two- and three-step ahead forecasting. The proposed model is compared with the non-denoising methods, namely the Bagging method, the partial least squares (PLS) model, the back propagation (BP) neural network, the support vector machine (SVM), the ELM and the AdaBoost-ELM models as well as several kinds of CEEMDAN-based and two-stage decomposition based methods. Results obtained from this study indicate that the proposed hybrid model can capture the nonlinear characteristics of wind speed time series and thus provide more accurate forecasting results compared with the other methods.