Abstract: The production pathway for cold-rolled silicon steel is relatively long, and the process control is complex. The production organization mode of conducting offline magnetic property testing during the final annealing stage of the finished product cannot meet the quality control requirements for correcting process deviations during intermediate stages to enhance product performance stability. In this paper, machine learning algorithms such as XGBoost, LightGBM, and Multi-Layer Perceptron (MLP) were utilized. By comparing the advantages and disadvantages of different algorithms, the magnetic property prediction models using XGBoost and LightGBM algorithms can meet the requirements for selective adoption and application in large-scale production conditions. These models support the prediction of the magnetic property levels of finished products during the production processes of various intermediate stages, thereby guiding process adjustments and ultimately stabilizing the magnetic properties of the final products.

Key words: silicon steel, magnetic property prediction, machine learning