Abstract: Based on the actual production data of steelmaking in a 260 t converter, the integrated learning algorithms of RF (Random Forests), LGBM (Light Gradient Boosting Machine) and Stacking integration were used to establish phosphorus and manganese prediction models for blowing endpoint in the converter. The model input variables were determined through the correlation theory analysis and Pearson correlation coefficient method. It was found by comparing the end-point hit rates of the three integrated learning models that the prediction performance of the Stacking integrated model is the best. With error tolerances of the phosphorus mass fraction at the endpoint of blowing being ±0.001 % and ±0.001 5 %, the hit rates are 86.3 % and 97.1 %, respectively. With error tolerances of the end point manganese mass fraction being ±0.008 %, and ±0.01 %, the hit rates are 83.4 % and 94.4 %, respectively.

Key words: converter blowing, endpoint phosphorus prediction, endpoint manganese prediction, machine learning, integrated algorithm, data driven model