Multi-step-ahead wind speed forecasting based on a hybrid decomposition method and temporal convolutional networks

Recently, the boom in wind power industry has called for the accurate and stable wind speed forecasting, on which reliable wind power generation systems depend heavily. Due to the intermittency and complexity of wind, an appropriate decomposition is proved as a pivotal part in the precise wind speed prediction. On this account, this paper constructs a hybrid decomposition method coupling the ensemble patch transform (EPT) and the complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN), where EPT is utilized to extract the trend of wind speed, then CEEMDAN is employed to divide the volatility into several fluctuation components with different frequency characteristics. Subsequently, the proposed decomposition method is combined with temporal convolutional networks (TCN) for the individual prediction of the trend and fluctuation components. Ultimately, the forecasted values for the wind speed prediction are obtained by reconstructing the prediction results of all the components. To evaluate the performance of the proposed EPT-CEEMDAN-TCN model, the historical wind speed data from three wind farms across China are used. The experimental results verify the notable effectiveness and necessity of the proposed EPT-CEEMDAN decomposition. In the meanwhile, the results demonstrate the significant superiority of the proposed EPT-CEEMDAN-TCN model on accuracy and stability. • The ensemble patch transformation (EPT) is established to extract the daily trend. • The CEEMDAN is applied to decompose the volatility of wind speed series. • The temporal convolutional networks (TCN) are adopted for prediction. • The EPT-CEEMDAN-TCN model is developed for multi-step-ahead wind speed forecasting. • The superiority of the proposed model is verified on datasets from diverse areas.