Deep Learning-Based Short-Term Load Forecasting in Smart Grids: Enhancing Energy Efficiency for Biotechnological Manufacturing and Commercialization

Abstract

Reliable and efficient energy management is essential for biotechnological manufacturing and commercialization, where continuous operations and stable power supply are critical. As electricity consumption grows, optimizing the load capacity of transmission lines emerges as the most economical solution to alleviate transmission pressure in smart grids. This study proposes a deep learning-based short-term load forecasting model designed to enhance energy efficiency and operational safety, particularly relevant for biotechnological facilities that require precise energy forecasting for uninterrupted production processes. The proposed forecasting model integrates a Feature Attention Mechanism (FA) with a Gated Recurrent Unit (GRU) to form the FAGRU prediction model, achieving reductions in Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) by 18.75% and 18.64%, respectively, compared to the conventional GRU model. By incorporating Wavelet Packet Decomposition (WPD), the WPD-FAGRU model demonstrated optimal performance at a time step of seven, achieving 69.11% and 69.17% reductions in MAE and RMSE, respectively. Furthermore, optimizing initial parameters using a Game Theory Strategy (GTS) resulted in the WPD-GTS-FAGRU model, which decreased the Mean Absolute Percentage Error (MAPE) by 13.96%, outperforming comparison models by significant margins. The high forecasting accuracy and improved operational safety demonstrated by the WPD-GTS-FAGRU model are of strategic significance for smart grid applications in the biotechnology sector. This model not only ensures reliable energy supply for sensitive biotechnological processes but also supports sustainable and cost-effective manufacturing practices, ultimately facilitating the efficient commercialization of biotechnological innovations.