基于Stacking模型融合的AOD终点碳温预测  

Prediction of end-point carbon content and temperature in AOD based on Stacking model fusion

作　　者：刘东旭 李明明 邵磊[1,2] 邹宗树 LIU Dongxu;LI Mingming;SHAO Lei;ZOU Zongshu(School of Metallurgy,Northeastern University,Shenyang 110819,China;Key Laboratory of Ecological Utilization of Multi-Metal Intergrown Ores,Ministry of Education,Northeastern University,Shenyang 110819,China)

机构地区：[1]东北大学冶金学院,辽宁沈阳110819 [2]东北大学多金属共生矿生态化冶金教育部重点实验室,辽宁沈阳110819

出　　处：《炼钢》2024年第4期30-39,共10页Steelmaking

基　　金：国家自然科学基金资助项目(51904062,52374329)。

摘　　要：为提高AOD不锈钢冶炼终点碳温预测的准确性和可靠性,提出一种基于多个机器学习算法(RF、XGBoost、AdaBoost、KNN、SVR和岭回归)嵌入的Stacking模型融合的AOD终点碳温预测方法。基于理论基础和相关性分析确定了模型的输入特征变量,利用箱线图法对历史数据进行预处理,结合5折交叉验证和贝叶斯优化算法确定了模型的最优参数,通过对上述6种机器学习算法的逐层筛选,构建了RF+XGBoost+KNN—RF二层Stacking多模型融合的终点碳含量预测模型、RF+AdaBoost+KNN—RF+XGBoost+KNN—XGBoost三层Stacking多模型融合的终点温度预测模型。预测结果表明,RF+XGBoost+KNN—RF二层Stacking模型在终点碳质量分数误差为±0.01%的命中率为87.86%,RF+AdaBoost+KNN—RF+XGBoost+KNN—XGBoost三层Stacking模型在终点温度误差为±15℃的命中率为94.22%,相较于单一的机器学习模型,预测精度显著提高。An end-point carbon content and temperature prediction method was proposed based on Stacking model fusion embedded in six machine learning algorithms(RF、XGBoost、AdaBoost、KNN、SVR and Ridge Regression),so as to improve the prediction accuracy and reliability of end-point carbon content and temperature in AOD.The input variables of the model were determined through fundamental theory related to AOD steelmaking and correlation analysis.The box plot method was used to preprocess the historical data,and the optimal model parameters were determined by combining the 5-fold cross-validation and Bayesian optimization algorithm.Consequently,a two-layer Stacking model(RF+XGBoost+KNN—RF)of end-point carbon content prediction and a three-layer Stacking model(RF+AdaBoost+KNN—RF+XGBoost+KNN—XGBoost)of end-point temperature prediction were established through screening the six machine learning algorithms.The prediction results show that the prediction accuracy of the RF+XGBoost+KNN—RF model is 87.86%within the carbon mass fraction error of±0.01%,and of the RF+AdaBoost+KNN—RF+XGBoost+KNN—XGBoost is 94.22%within the temperature error of±15℃,which are significantly improved compared with the single machine learning model.

关 键 词：AOD 终点碳温预测 模型融合 Stacking模型 

分 类 号：TF769.1[冶金工程—钢铁冶金]