Ultra-short-term wind speed forecasting based on secondary decomposition and Transformer-MLR combined model
Yong Yue, Weiming Zheng, Anguo Wu, Xin Jin, Zhongquan Huang, Hongsheng Zhang
Abstract
Ultra-short-term wind speed forecasting is critical for optimizing wind turbine control and real-time power grid load distribution. This study proposes a novel multivariate time series forecasting approach that combines a secondary decomposition strategy with a hybrid prediction model to enhance forecasting accuracy. In the decomposition phase, the original wind speed data is first processed using adaptive complete ensemble empirical mode decomposition. Subsequently, variational mode decomposition is applied to the high-frequency sub-sequences for further decomposition, effectively reducing the series' non-linearity. In the prediction phase, the Transformer model is employed to capture the periodic characteristics of high-frequency sub-sequences, while medium- and low-frequency sub-sequences are modeled using multivariate linear regression. The final wind speed forecasting is obtained by integrating the forecasts of all sub-sequences. Real wind farm data is used for model validation, and the performance is evaluated with four metrics: root mean squared error (RMSE, 0.2260), mean absolute error (MAE, 0.1736), mean absolute percentage error (MAPE, 2.51 %), and R-squared (R 2 , 0.9964). The results demonstrate that the proposed model achieves superior forecasting accuracy, supporting improved wind farm operational efficiency and power grid reliability.