转炉终渣成分和溅渣时间数据驱动预测模型开发  

Development of data driven prediction model for endpoint slag composition and slag splashing time of converter

作　　者：洪科 赵自鑫[2] 钟良才 于海岐[2] 刘承军 HONG Ke;ZHAO Zixin;ZHONG Liangcai;YU Haiqi;LIU Chengjun(School of Metallurgy,Northeastern University,Shenyang 110819,China;Bayuquan Iron and Steel Company,Angang Steel Co.,Ltd.,Yingkou 115000,China)

机构地区：[1]东北大学冶金学院,沈阳110819 [2]鞍钢股份有限公司鲅鱼圈钢铁分公司,辽宁营口115000

出　　处：《炼钢》2023年第4期21-27,共7页Steelmaking

基　　金：中央高校基本科研业务专项资金资助项目(N2125018);国家自然科学基金资助项目(51574069);科技部国家重点研发计划资助项目(2017YFB0304100)。

摘　　要：基于260 t转炉实际生产数据,通过机器学习算法XGBoost(eXtreme Gradient Boosting,极限梯度提升树)、弹性回归、线性回归、AdaBoost(Adaptive Boosting,自适应提升树)四种算法建立了终渣主要成分CaO、SiO_(2)、TFe和MgO的预测模型。通过优化调参,XGBoost终渣成分预测模型的决定系数R^(2)均在0.8以上。溅渣时间模型采用SVR(Support Vector Regression,支持向量机回归)、LGBM(Light Gradient Boosting Machine,轻量梯度提升机回归)、GBDT(Gradient Boosting Decision Tree,梯度提升树回归)、RF(Random Forest,随机森林)和XGBoost五种算法进行建模。通过探究,将SVR、XGBoost、GBDT算法使用集成方法得到Stacking集成溅渣时间预测模型,Stacking集成溅渣时间预测模型提升了单个模型的预测效果,偏差为±20 s的预测命中率达89.95%。Based on the actual production data of 260 t converter,prediction models of the main components of endpoint slag,CaO,SiO_(2),TFe and MgO,were established through four algorithms of machine learning algorithm,XGBoost(eXtreme Gradient Boosting),elastic regression,linear regression and AdaBoost(Adaptive Boosting).By optimizing the parameters,the determination coefficients R^(2) of XGBoost endpoint slag composition prediction model is all above 0.8.The slag splashing time model was modeled using five algorithms:SVR(Support Vector Regression),LGBM(Light Gradient Boosting Machine),GBDT(Gradient Boosting Decision Tree),RF(Random Forest)and XGBoost,respectively.Then,the integrated slag splashing time model was obtained by integrating SVR,XGBoost and GBDT.Stacking integrated slag splashing time model improves the prediction effect of each single model,and the prediction hit rate is 89.95%in the error range of±20 s.

关 键 词：转炉 溅渣护炉 终渣成分 溅渣时间 预测模型 机器学习 

分 类 号：TF711[冶金工程—钢铁冶金]