Feature Selection and Prediction of Mechanical Properties of Steel Coils Based on Gradient Boosting Decision Tree

The basic modeling features of hot-dip galvanized steel coil were selected on the basis of metallurgical mechanism, and then other chemical element features were selected by gradient boosting decision tree algorithm. The model parameters were optimized by combining grid search with cross validation methods, and the effects of different features on yield strength of steel coils were analyzed by the model. The results show that the basic modeling features of mechanical property prediction model of hot-dip galvanized steel coils included process parameter features, specification features and basic chemical element features, and other chemical element features which had great influence on the yield strength of steel coils were nitrogen and aluminum content. After optimizing the model parameters, the root mean squared error of the yield strength measured on the test set was 10.671 MPa, the mean absolute error was 8.244 MPa, and the mean absolute percentage error was 2.641%. The prediction accuracy of the model was significantly higher than that before optimizing the model parameters. When the content of carbon, silicon or manganese or the hot rolling in-rolling temperature changed, the yield strength of steel coils changed greatly.