基于异质堆叠集成学习的脱硫剂加入量预测  

Prediction of desulfurizer addition based on heterogeneous stack ensemble learning

作　　者：吴经纬[1] 方一飞 但斌斌[2] 容芷君[2] 都李平 罗钟邱 WU Jingwei;FANG Yifei;DAN Binbin;RONG Zhijun;DU Liping;LUO Zhongqiu(WISDRI Engineering&Research Incorporation Limited,Wuhan 430223,China;Key Laboratory of Metallurgical Equipment and Control,Ministry of Education,Wuhan University of Science and Technology,Wuhan 430081,China;Baoxin Software(Wuhan)Co.,Ltd.,Wuhan 430080,China)

机构地区：[1]中冶南方工程技术有限公司,湖北武汉430223 [2]武汉科技大学冶金装备及其控制教育部重点实验室,湖北武汉430081 [3]宝信软件(武汉)有限公司,湖北武汉430080

出　　处：《炼钢》2024年第3期1-7,共7页Steelmaking

基　　金：国家自然科学基金资助项目(51475340);湖北省重点研发计划项目(YFXM2022000556);湖北省中央引导地方科技发展资金项目(2020ZYYD022)。

摘　　要：KR搅拌法是铁水预脱硫的重要工艺之一,但在脱硫过程中脱硫剂的加入量主要依赖于人工经验控制,导致一次脱硫命中率较低,铸坯质量不稳定。为准确控制脱硫剂加入量,提高一次脱硫命中率,提出一种基于异质堆叠集成学习的脱硫剂加入量预测方法。首先,进行数据清洗,采用LOF算法结合专家经验剔除异常值。其次,采用最大信息系数结合斯皮尔曼秩相关系数进行特征筛选。最后,引入堆叠集成算法,基于评价指标和误差相关性分析优选模型的不同基学习器,并采用Optuna算法为基学习器寻找最优参数组合,建立异质堆叠集成预测模型。以现场采集的脱硫数据作为样本进行实例分析,结果表明,模型的决定系数(R^(2))为91.6%,均方根误差(RMSE)为197.79,平均绝对误差(MAE)为117.14,平均绝对百分比误差(MAPE)为6.95%。KR mixing method is one of the important processes for pre-desulfurization of hot metal,but the amount of desulfurizer added in the process mainly depends on manual experience control,which leads to low hit rate of desulfurization and unstable quality of the billet.In order to accurately control the amount of desulfurizer added and improve the hit rate of primary desulfurization,a method for predicting the amount of desulfurizer added based on heterogeneous stack ensemble learning was proposed.First of all,the data was cleaned and the outliers were eliminated using LOF algorithm combined with expert experience.Secondly,the maximum information coefficient combined with Spearman rank correlation coefficient was used for feature screening.Finally,the stack integration algorithm was introduced to select different base learners of the model based on the evaluation index and error correlation analysis,and Optuna algorithm was used to find the optimal parameter combination for the base learner,and the heterogeneous stack integration prediction model was established.The results show that the coefficient of determination(R^(2))is 91.6%,the root mean square error(RMSE)is 197.79,the mean absolute error(MAE)is 117.14,and the mean absolute percentage error(MAPE)is 6.95%.

关 键 词：KR搅拌法 脱硫剂加入量 特征筛选 Optuna 堆叠集成 

分 类 号：TF704.3[冶金工程—钢铁冶金]