The steelmaking process involves the removal of impurities from molten iron to achieve the desired material properties and ensure the quality of the final steel product. Proper control of molten steel temperature is crucial because it directly and indirectly impacts product quality. Inadequate temperature control can lead to defects in the final product or failure to meet the desired material specification. Traditionally, machine learning algorithms have been applied to predict molten steel temperature by treating each steelmaking cycle in the converter as an independent event. However, such simple regression-based approaches often fail to capture the sequential characteristics of the converter process. In this study, we propose a time-series modeling approach that incorporates residual heat information from previous cycles to better reflect the continuous operational characteristics of the converter. This method enables the identification of operational patterns that conventional models overlook and demonstrates improved temperature prediction accuracy. The proposed predictive model is expected to contribute to future optimization of raw material and energy inputs required to achieve the target temperature.
Han et al. (Sat,) studied this question.