Short-Term Electricity Forecasting of Rolling Shops of Metallurgical Enterprises

In modern metallurgy, energy efficiency and optimization of electricity use have become critical factors in the competitiveness of manufacturing companies. The increasing energy costs and requirements for environmental compliance encourage companies to implement innovative technologies for energy forecasting and management. Special attention is given to short-term hourly energy consumption forecasts, which help optimize purchases on the wholesale electricity market and reduce costs. The production of hot and cold rolled steel products, which are among the most energy-intensive stages of the metallurgical process, requires effective energy management through accurate forecasting models that consider the specifics of production processes and factors. This article explores machine learning techniques for predicting hourly energy consumption in sheet metal processing one day in advance. The research was conducted by analyzing production data over the past 8 years. Models have been developed and tested based on architectures Random Forest and LSTM, as well as compared them to an existing one-factor regression model. The LSTM model with bidirectional layer showed the best results.