<p>High-purity aluminum, possessing unique advantages such as ultra-low impurity depth, superior electrical and thermal conductivity, and excellent corrosion resistance, has become an indispensable material in advanced manufacturing. However, the process of producing high-purity aluminum through electrolysis involves complex interactions among process parameters. Under the demand for low carbon emissions, predicting and optimizing energy consumption are particularly important. This study develops a machine learning framework for energy consumption modeling in high-purity aluminum production. Operational data from electrolytic cell control systems were combined with manual production records, forming a comprehensive parameter sets of 3114 reliable data. Through Pearson correlation analysis and rigorous data preprocessing, key process variables were identified. Ten machine learning algorithms were systematically evaluated, with LightGBM demonstrating superior predictive performance (<i>R</i>² &gt; 0.9 on test dataset). Feature importance analysis guided model optimization, revealing critical process parameters affecting energy efficiency. The optimized LightGBM model was implemented in actual production scenarios, enabling data-driven process adjustments that significantly improved the capacity of high-purity aluminum production. This approach provides an effective methodology for intelligent optimization of energy-intensive industrial processes, demonstrating the practical value of machine learning in metallurgical manufacturing.</p> Graphical Abstract <p></p>