Temperature prediction model for ladle furnace based on mathematical mechanisms and the GA–BP algorithm

Abstract

This study addresses the prediction inaccuracy and poor adaptability of conventional temperature prediction models for ladle furnace refining. The historical production data of a steel plant were used to establish a hybrid prediction model based on mathematical mechanisms and a backpropagation neural network optimized using a genetic algorithm. The coefficient of determination ( R 2) of the hybrid model was 0.98, and the hit ratio of temperature prediction within ±5°C was 99.3%. The ladle's thermal status affected the model prediction accuracy. A steel plant with a compact production rhythm and good baking state was less affected by the ladle's thermal status. The model input variables exhibited varying degrees of influence on the end temperature of molten steel in two steel plants with the starting temperature of the molten steel entering the station having the greatest influence. For accurate temperature control and prediction in actual production, high-influence variables must be given focus to ensure their stable control and process interference must be mitigated.