The energy sector’s rapid expansion necessitates accurate, dependable, and computationally efficient short‐term load forecasting (STLF) models to assure real‐time balance between energy supply and demand. However, the stochastic nature of the energy usage and its reliance on changing weather conditions make accurate forecasting difficult. This paper presents an innovative deep learning‐based STLF architecture for both residential and commercial applications, which tackles these constraints with three significant innovations. First, it proposes a simple yet efficient data imputation strategy that improves model robustness by handling missing or noisy data. Second, it has a series core fusion (SCF) method in conjunction with a star aggregate‐redistribute (STAR) module. Unlike traditional attention methods, which rely on scattered inter‐channel interactions, STAR centralizes information aggregation, lowering computing overhead and reducing reliance on individual channel quality, making it a more effective substitute for regular attention layers. Third, an improved particle swarm optimization (IPSO) technique is used to automatically adjust hyperparameters, resulting in an optimal model setup without manual intervention. The proposed model generates minute‐level predictions and refines them with a day‐type categorization technique (weekday, weekend, holiday). When tested on three real‐world benchmark datasets, the proposed framework outperformed state‐of‐the‐art (SOTA) models, lowering root mean square error (RMSE) by 59.41%, mean absolute error (MAE) by 30.58%, and mean absolute percentage error (MAPE) by 12.5%. Furthermore, the proposed model’s low computational requirements make it suitable for real‐time implementation on edge devices. These contributions provide a scalable and economical solution for smart grid operation, microgrid control, and demand‐side energy management, therefore advancing the practical application of intelligent forecasting systems in current power systems.
Iqbal et al. (Wed,) studied this question.