Optimized Deep Learning Architecture for Short-Term Wind Speed Estimation

Wind energy is a sustainable energy source that generates clean, pollution-free power. Accurate wind speed prediction is critical for maximizing wind energy generation, increasing grid stability, and simplifying energy trade. The proposed model investigates the use of DNN for wind speed forecasting, using historical wind data to train prediction models. To improve model prediction, this study uses several data preprocessing strategies such as normalization, feature selection, and missing value handling. The suggested DNN model is trained and verified on wind speed information collected from NREL, India and by tuning parameters such as neurons, hidden layers, and activation functions and optimization techniques. The proposed model is estimated using evaluation metrics, along with three input features and a consistent training procedure. Among all the models 3-layer DNN with Swish activation, 4 neurons per layer, and the Adam optimizer, the best performance was achieved with low MAE 2.04 and MSE of 7.12. These findings emphasize the potential of DNN-based models for better wind energy forecasting, ultimately leading to more sustainable energy scheduling and safe power grid operations.