PREDICTION AND ANALYSIS OF INDUSTRIAL ENERGY CONSUMPTION USING MACHINE LEARNING TECHNIQUES

Abstract

Energy use is a critical issue for efficiency and sustainable production in particular industries where energy use is high. Predicting and understanding energy demand can aid energy monitoring, planning and efficiency. This study aimed to analyze and predict industrial energy consumption using machine learning techniques while identifying the key variables influencing energy demand. A quantitative secondary data analysis was conducted using 35,040 observations of electrical, temporal, and operational variables from an industrial steel production context. CO₂ emissions were excluded to avoid redundancy. Descriptive statistics, correlation analysis, and temporal trend evaluation were performed. Three regression models (Linear Regression, Decision Tree and Random Forest) were trained on an 80:20 split. We used MAE, RMSE and R² to evaluate the model and the Random Forest feature importance to determine important features. Energy consumption exhibited a right-skewed distribution with clear hourly operational patterns. The Random Forest model achieved the best performance, with MAE = 0.2917, RMSE = 0.8864, and R² = 0.9993. Lagging reactive power emerged as the dominant predictor, accounting for 87.03% of feature importance, followed by lagging power factor. These findings demonstrate that machine learning models can effectively predict industrial energy consumption while providing interpretable insights into key electrical drivers. The results highlight the importance of reactive power and power factor management for improving industrial energy efficiency.