Feature Engineering to Embed Process Knowledge: Analyzing the Energy Efficiency of Electric Arc Furnace Steelmaking

The importance of electric arc furnace (EAF) steelmaking is expected to increase worldwide as parts of the industry transition to lower carbon dioxide emissions. This work analyzed one year’s operational data from an EAF plant that uses a large proportion of direct-reduced iron (DRI) in the furnace feed. The data were used to test different approaches to quantifying the effects of process conditions on specific electricity consumption (kWh per ton of crude steel). Previous work indicated that inputs such as the proportion of DRI, fluxes, natural gas, and oxygen were linearly correlated with specific electricity consumption. The current study confirmed that conventional multiple linear regression (MLR) can reproduce electricity consumption trends in EAF steelmaking; however, many model coefficients deviated significantly from expected values. Implementing engineered features—the slag volume and total carbon input—yielded coefficients more aligned with expectations, though prediction accuracy did not improve. An enhanced approach using engineered features in a non-linear machine learning model (XGBoost) resulted in both physically reasonable trends and lower prediction errors. The analysis from Shapley dependence consistent with theoretical trends further validated the model's performance.