Optimization-Driven Framework for High-Precision Wind Speed Forecasting: An Investigation of Advanced Computational Models

For wind energy systems to operate, plan, and be reliable, accurate wind speed forecasts are essential. Conventional methods frequently have low accuracy or processing inefficiency, which limits their use in real-time situations. this study proposes a Multi Layer Perceptron (MLP) -based Deep Learning (DL) model that is optimized using the Northern Goshawk Optimization Algorithm (NGOA). Unlike existing work, the model utilizes merely meteorological and temporal features, wind speed, wind direction, turbulent kinetic energy, and calendar-based variables. It employed a Florida subset of the NREL WIND Toolkit of sequential instances with a 75% − 25% sequential split to preserve temporal order and prevent data leakage. Six models—basic MLP, deep MLP, wide MLP, shallow MLP, dropout MLP, and deep-dropout MLP-were trained and evaluated. Of all models that were compared, the NGOA -deep MLP performed the best, recording the lowest test errors and highest explanatory power, achieving a Coefficient of determination (R²) of 0. 9530 in training, and 0. 9082 in testing, also Explained Variance (EV) of 0. 9532. Interestingly, turbulent kinetic energy and diurnal cycles (hour of day) were the best predictors, according to Shapley Additive Explanation. With 3-7 seconds for train times and 1-2 seconds for test times, the NGOA-tuned deep MLP architecture is both accurate and computationally efficient, making it practical for real-time wind speed forecasting. Its application can enhance grid stability, energy scheduling, and cost-efficient management in renewable wind power systems.