关键词: 转炉炼钢, 终点磷含量, 集成学习, Stacking模型

Abstract: Accurate prediction of endpoint phosphorus content can solve the problem that the composition of molten steel cannot be continuously detected during the converter blowing process, promote the rapid tapping of the converter and shorten the smelting cycle.Based on the actual production data of a 100 t converter in a steel mill in Hunan Province, a prediction model of molten steel phosphorus content at the end of the converter based on an ensemble learning algorithm was established. The main influencing factors were determined by analyzing the converter dephosphorization process, and the input variables of the model were determined based on the actual production of the steel mill. The interquartile range method, LOF algorithm and other methods were used for data preprocessing, and the best parameters of the model were determined by combining the 10-fold cross validation method and the grid search method.Four ensemble learning algorithms (RF, AdaBoost, MultiBoosting, and Stacking) were used to establish the models, and  the prediction results of the four models were compared, it was found that the Stacking model had the best prediction effect, with the highest hit rate, the lowest root mean square error and the average absolute percentage error, and the hit rates were 86.08% and 97.84% when the error of molten steel phosphorus mass fraction at the prediction end point was ±0.003% and ±0.005%, respectively.

Key words: converter steelmaking, endpoint phosphoruscontent, ensemble learning, Stacking model