Regional solar generation prediction with metaheuristically optimized artificial intelligence for sustainable grid management

The growing reliance on solar energy in modern power grids demands accurate forecasting to ensure stability, reliability, and efficient energy management. However, the inherently unpredictable nature of solar power—driven by variable weather conditions—makes precise prediction challenging. Photovoltaic (PV) generation data also exhibit trends, seasonal variations, and noise, while excessive meteorological inputs can introduce redundancy and degrade model performance. This study proposes a hybrid AI-based forecasting framework that integrates wavelet transform, correlation-based feature engineering, and metaheuristic optimization to enhance short-term solar forecasting. The wavelet transform extracts multi-resolution features from raw PV signals, while correlation analysis identifies the most relevant meteorological variables. To minimize redundancy, embedded feature selection retains only the most informative predictors. These features are then used to train advanced machine learning and deep learning models. The top-performing model is further fine-tuned using the Jellyfish Search algorithm, which streamlines hyperparameter optimization while enhancing accuracy and robustness. Model performance is evaluated using multiple error metrics and a comprehensive validation index. In addition, a user-friendly graphical interface has been developed to facilitate adoption by power companies and grid operators. Results demonstrate that combining wavelet decomposition, feature engineering, and metaheuristic optimization with AI models significantly improves the accuracy and reliability of short-term solar forecasting.