Steel Manufacturing Energy Consumption Forecasting for Decision Support Using Initial Operational Data

The steel manufacturing industry is highly energy-intensive, making accurate estimation of future energy consumption important for operational planning and energy management. Traditional energy monitoring mainly focuses on historical consumption and may not provide sufficient information about future requirements. This study investigates whether initial operational data can be used to forecast future energy consumption over an 8-hour horizon in a steel manufacturing environment. The study uses the DAEWOO Steel industry energy consumption dataset, recorded at 15-minutes intervals. Exploratory Data Analysis (EDA) is performed to examine the structure, distributions, relationships and temporal patterns of the data After preprocessing And feature engineering, a finalised set of 17 operational and temporal features is used for forecasting. A chronological 80 / 20 train-test separation is applied to preserve temporal order and reduce information leakage. A Decision Tree Regressor is considered as the research baseline, while Random Forest and a Deep Feedforward Network (DFFN) are evaluated as predictive models. The DFFN is selected as the primary forecasting model based on its stronger overall-performance, achieving an RMSE of 368.1558 kWh and R of 0.7322 on the test set, while Random Forest achieved an RMSE of 393.2960 kWh and R 0.6944. The framework further incorporates local explainability,an RF Supporting Perspective, What-If Analysis, and Decision-Support interpretation of estimated energy and cost changes. These components are integrated into SteelSense,a web-based forecasting application. The study supports data-driven forecasting for energy management and decision making.