Prediction Model of Liquid Level Fluctuation in Continuous Casting Mold Based on GA-CNN

Abstract

In this paper, a neural network model is introduced, which uses Convolutional Neural Network model optimized by Genetic Algorithm (GA-CNN) to predict the liquid level fluctuation of continuous casting mold in real time. The experimental data were obtained from the practical production process of an iron and steel plant in China. For the two-stream continuous caster in this factory, two sets of mold liquid level fluctuation value data sets were established. These data sets are categorized into the first- and the second-stream mold liquid surface fluctuation data set. Both data sets comprise 138 production parameters along with the mold liquid level fluctuation values. Random Forest feature screening and data preprocessing are carried out on the data set, so that the processed data can be learned and trained by the model, thus obtaining the mold liquid level fluctuation prediction model. To facilitate the analysis, the sensitivity analysis of related continuous casting production parameters and mold level fluctuation data was carried out. Subsequently, the influence of production parameters on the fluctuation value of mold liquid level was investigated by characteristic thermal diagram and Shapley additional explanations diagram, and the influence degree of each parameter on the model performance was determined. The results indicate that the GA-CNN model demonstrates strong predictive capability, with a high correlation coefficient (R²) of 0.98, a low mean absolute error (MAE) of 0.1093, and a mean square error (MSE) of 0.0225. Notably, the position of the stopper was identified as a critical factor significantly affecting mold level fluctuations. The model has high prediction accuracy and can meet the needs of practical application in steel plants.