VMD-GRU Based Ultra-Short-Term Wind Speed Forecasting System

This paper proposes a solution for ultra-short-term wind speed forecasting over horizons of 10, 20 and 30 minutes using real SCADA data from a wind turbine. The proposed approach combines Variational Mode Decomposition (VMD) with a Gated Recurrent Unit (GRU) neural network in a hybrid forecasting framework. VMD is first applied to decompose the original wind speed signal into several intrinsic mode functions (IMFs) that separate low- and high-frequency variations and reduce the non-stationary behaviour of the series. The selected IMFs are then used as inputs to the GRU model in order to better capture short-term temporal dependencies and rapid wind fluctuations. The proposed VMD-GRU model is evaluated using SCADA measurements and compared with conventional deep learning models including CNN, LSTM and GRU trained directly on the raw data. Performance is assessed using RMSE, MAE and the coefficient of determination$\mathbf{R}^{2}$. The results show that the hybrid model achieves lower RMSE and MAE and higher R2across all three horizons, demonstrating its effectiveness and robustness for ultra-short-term wind speed forecasting.