Prediction of Mechanical Properties of Hot-Rolled Strip Steel Based on XGBoost and Metallurgical Mechanism

High-quality data and high-performance models are two critical factors affecting the accuracy of predicting the mechanical properties of strip steel. This study aims to construct a model for predicting the mechanical properties of hot-rolled strip steel with high generalizability and reliability based on metallurgical mechanisms and deep neural networks. Data features at different scales are extracted and ranked in importance to filter out the features affecting the mechanical properties of the steel strips by mechanism analysis. Three neural network prediction models based on LR, LS-SVM, and XGBoost are constructed, and the hyper-parameters of the three models are tuned using the grid search method to achieve the best performance. The prediction results of the three models show that when the prediction error is less than 30 MPa, the accuracy of the mechanical properties prediction model of hot rolled strip steel constructed based on XGBoost reaches 97.4%; when the prediction error is less than 50 MPa, the accuracy reaches 99.4%. The model’s accuracy is higher than other models and has good generalization.