Wind Power Forecasting Using Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU) and Hybrid Model

Wind power forecasting is decisive for the incorporation of wind energy into the power grid, as it enables wind farms to optimize their operational efficiency and contributes to maintaining grid stability. This study evaluates three deep learning models namely Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), and a Hybrid Model that integrates both architectures—for forecasting short-term wind power at the London Array wind farm, one of the largest wind farms in Europe. The LSTM and GRU models achieved root mean square errors (RMSEs) of 43.90 and 44.06, respectively, whereas the Hybrid model demonstrated a significant improvement with an RMSE of 38.28. Regarding mean absolute error (MAE), the LSTM model recorded a value of 26.32, the GRU model 25.60, and the Hybrid model further enhanced performance with an MAE of 21.90. The R² metric was recorded at 0.97 for LSTM and 0.96 for GRU models, while the Hybrid model attained an R² value of 0.97, indicating a strong fit for all models in capturing the variance in wind power data. These results demonstrate the efficacy of these deep learning models in precise wind power forecasting, offering benefits like enhanced grid stability, economic advantages through efficient energy market participation, and potential reductions in consumer energy prices, highlighting the models' applicability in the renewable energy sector.