Enhancing steelmaking productivity: a hybrid approach of machine learning and genetic algorithm optimisation for electric arc furnace

Abstract

This study presents a hybrid optimisation approach to enhance electric arc furnace (EAF) operations by minimising arcing time through the integration of machine learning models and genetic algorithm (GA) optimisation. The goal is to improve productivity, reduce energy consumption and lower operational costs. A comprehensive historical dataset of EAF operations was utilised to develop predictive models, including Gaussian process regression (GPR), decision trees, support vector regression, and artificial neural networks for predicting optimal arcing time. Key input parameters such as raw material quality, direct reduced iron composition, power input, oxygen and natural gas usage, carbon, flux additions and furnace temperature were considered. Feature engineering technique R-Relief was used to identify significant variables, and hyperparameter tuning improved model performance. The exponential GPR model achieved coefficient of determination (R²) of 0.93 and root mean square error of 0.75. GA optimisation reduced arcing time by 5%, increasing steel production from 216.6 to 223 tons per hour.