A data-driven multi-objective differential evolution framework for charging-thermal coupled parameter optimization in rotary hearth furnace operations

The rotary hearth furnace (RHF) is a high-temperature reactor used for metallurgical solid waste recycling and is characterized by multi-phase and multi-physics interactions. The inherent complexities of these couplings, together with conflicting objectives such as metal yield and pollutant emissions, pose significant challenges for operational optimization of charging-thermal coupled parameters in the RHF. To address these challenges, this article proposes a data-driven multi-objective differential evolution (MODE) framework: (1) an Extreme Gradient Boosting (XGBoost) surrogate model is developed to enable accurate prediction of pollutant emissions and metal yield; and (2) a self-adaptive MODE (SaMODE) algorithm with S-metric selection and Gaussian local search is proposed to simultaneously optimize emissions and yield. Experimental results demonstrate that (1) the XGBoost surrogate model achieves superior prediction accuracy and computational efficiency compared to other machine learning models; and (2) the proposed SaMODE algorithm obtains a high-quality Pareto optimal solution set (POS) and outperforms other multi-objective evolutionary algorithms. Furthermore, an entropy weighting method is employed to select a solution from the POS, resulting in simultaneous reduction of pollutant emissions and improvement of metal yield in an industrial case study.