Jellyfish Search-optimized wavelet transform-enhanced artificial intelligence for multi-horizon energy consumption forecasting in the Mid-Atlantic region

Accurate forecasting of energy consumption is critical for operational planning, maintenance, sustainability initiatives, economic projections, and infrastructure development. This study proposes a novel hybrid model that integrates artificial intelligence (AI) techniques, wavelet transforms, and metaheuristic optimization to predict energy usage across multiple time horizons in the Mid-Atlantic region of the United States. The region's complex urban dynamics, climate variability, and integration of renewable energy present significant forecasting challenges. To address these, the wavelet transform is employed to manage trends, noise, and non-linearity. At the same time, the bio-inspired Jellyfish Search (JS) optimizer fine-tunes model hyperparameters to reduce prediction errors. The integration of wavelet transforms enhances the performance of AI models—including convolutional neural networks (CNNs), time-series deep learning architectures, and ensemble machine learning methods—achieving mean absolute percentage error (MAPE) reductions of up to 36.33 %. The best-performing models attain MAPE values of 0.73 %, 5.59 %, 7.74 %, 7.66 %, and 7.97 % for 1-h, one-day, one-week, one-month, and one-year forecasts, respectively. To the authors' knowledge, this is the first study to combine AI, wavelet transforms, and metaheuristic algorithms for forecasting regional short- and long-term energy consumption, thereby substantially improving model robustness and predictive accuracy.