Speaker
Description
Forecasting electricity consumption in industrial enterprises is essential for cost minimization and effective participation in balancing markets, particularly in the context of continuous production processes. The liberalization of energy markets and the increasing share of renewable energy sources further complicate the planning of electricity demand. This study aims to perform a comparative evaluation of methods for short-term electricity consumption forecasting in an industrial facility equipped with metal-cutting machines operating in a two-shift regime and characterized by a nonlinear daily and weekly load structure. A total of nine methods are analyzed, including statistical models, machine learning techniques, and deep neural networks, as well as a hybrid XGBoost+LSTM model, under unified experimental conditions. The results indicate that LSTM achieves the highest accuracy (MAE ≈ 3.5 kWh, R² ≈ 0.996), due to its ability to capture long-term temporal dependencies. Ensemble methods (Random Forest and XGBoost) also demonstrate high accuracy (MAE ≈ 5 kWh) with lower computational requirements, making them suitable for a wide range of industrial applications. Based on the obtained results, practical guidelines are formulated for selecting an appropriate forecasting model depending on data availability, computational resources, and operational requirements.