Enhancing Electricity Demand Forecasting Based on a Hybrid Deep‐Learning Framework

Accurate electricity demand forecasting is crucial for the stable operation of the power system. A hybrid electricity demand forecasting framework, CVS‐iLSTNet, was proposed, which integrates Complete Ensemble Empirical Mode Decomposition (CEEMDAN), Variational Mode Decomposition (VMD), Sparrow Search Algorithm (SSA), and Long Short‐Term Network (LSTNet) to enhance forecasting accuracy and robustness. First, feature selection combining Recursive Feature Elimination (RFE) and Random Forest (RF) analyzes the relationship between electricity demand and influencing factors. Second, the Kolmogorov‐Arnold Network (KAN) layer is introduced to improve its nonlinear representation capabilities. Next, CEEMDAN decomposes the raw time series data to extract intrinsic modal functions (IMFs). VMD further decomposes these IMFs to extract finer modes, which are then combined with the corresponding IMFs. Finally, SSA‐iLSTNet enhances the prediction of complex patterns by hyperparameter optimization and combining multiple network structures and algorithms. Experiments were conducted using real electricity demand data from Spain and the United States. The results demonstrate that the proposed model delivers reliable and timely predictions both in short‐term and long‐term demand forecasting, outperforming state‐of‐the‐art models. © 2025 Institute of Electrical Engineers of Japan and Wiley Periodicals LLC.