Knowledge model for response to ladle furnace refining time disturbance based on endpoint molten-steel temperature prediction

Ladle furnace (LF) refining is a critical process in the steelmaking–continuous casting (SCC) process, in which the end-point temperature directly determines the continuity of molten-steel casting at the caster. Owing to the strong uncertainty inherent in the LF refining process, deviations between the actual and planned operation times of LF refining (LF refining time disturbance) frequently occur when the desired endpoint temperature is targeted, which hinder the smooth operation of the SCC process. To address this issue, this study first established an endpoint molten-steel temperature prediction model based on CatBoost and optimized by a grid search (GS-CatBoost), which was used to predict the endpoint molten-steel temperature under the planned production parameters. Second, a knowledge model was constructed to determine the response to a LF refining time disturbance. Based on the predicted endpoint temperature, the degree of deviation between the actual and scheduled operation times was calculated, and the corresponding countermeasures were proposed with the support of an LF refining time disturbance response knowledge graph. Furthermore, the proposed endpoint temperature prediction model was trained and tested using historical industrial data from a domestic steel plant. The hit ratio reached 90.99% within an error range of ±5°C, with a coefficient of determination of 0.9202 and root mean squared error of 3.1231. Based on the prediction of the endpoint molten-steel temperature, the corresponding process operations and examples of response strategies for different levels of time disturbance were proposed. In addition, the countermeasures generated by the LF refining time-disturbance response knowledge model could be further integrated with scheduling algorithms for practical industrial applications.