A Meta-Learning-Based Approach for Predicting Steel Rolling Thickness Under Roll and Specification Changes

To address the issue of low control accuracy in the head thickness of the first coil after roll change and specification change in the hot-rolled strip steel production process, this paper proposes a feature extraction method based on hybrid target encoding, singular value decomposition (SVD), and SHAP. Additionally, a data-driven model based on meta-learning is established, which predicts the head thickness of the first coil after roll change and specification change using Model-Agnostic MetaLearning (MAML). The experimental results show that the proposed model achieves higher prediction accuracy and stability compared to existing models, and is more adaptable to operational condition changes and small sample scenarios. This provides new ideas and methods for thickness control in the hot-rolled strip steel production process.