A Hybrid Neural Network Model for Short-Term Wind Speed Forecasting

This study proposes an effective wind speed forecasting model combining a data processing strategy, neural network predictor, and parameter optimization method. (a) Variational mode decomposition (VMD) is adopted to decompose the into multiple subseries where each contains unique local characteristics, all are converted two-dimensional samples. (b) A gated recurrent unit (GRU) sequentially modeled based on obtained samples makes predictions for future speed. (c) The grid search with rolling cross-validation (GSRCV) designed simultaneously optimize key parameters of VMD GRU. To evaluate effectiveness proposed VMD-GRU-GSRCV model, comparative experiments hourly collected from National Renewable Energy Laboratory implemented. Numerical results show that root mean square error, absolute percentage symmetric error this reach 0.2047, 0.1435, 3.77%, 3.74%, respectively, which outperform benchmark using popular methods, techniques, hybrid models.