一种基于即时学习局部模型的LF终点预测方法  

An end-point prediction method for LF refining based on just-in-time learning local model

作　　者：楚建伟 刘建华[1] 何杨[1] 许庆礽 尤大利 罗仁辉[4] CHU Jianwei;LIU Jianhua;HE Yang;XU Qingreng;YOU Dali;LUO Renhui(National Engineering Research Center for Advanced Rolling and Intelligent Manufacturing,University of Science and Technology Beijing,Beijing 100083,China;School of Advanced Engineering,University of Science and Technology Beijing,Beijing 100083,China;Ferrous Metallurgy,Montanuniversitaet Leoben,Leoben 8700,Austria;Xinyu Iron and Steel Co.,Ltd.,Xinyu 338001,China)

机构地区：[1]北京科技大学高效轧制与智能制造国家工程研究中心,北京100083 [2]北京科技大学高等工程师学院,北京100083 [3]莱奥本矿业大学钢铁冶金系,奥地利莱奥本8700 [4]新余钢铁集团有限公司,江西新余338001

出　　处：《冶金自动化》2023年第1期147-155,共9页Metallurgical Industry Automation

基　　金：国家重点研发计划政府间国际科技创新合作(2021YFE0113200)。

摘　　要：LF精炼是连接转炉和连铸的重要工序,对其终点进行精准预测有助于提高LF生产效率,确保后续工序稳定进行。为提高LF终点预测准确率,提出了一种基于即时学习局部模型的LF精炼终点预测方法。在即时学习框架下,采用特征重要性和时间双重加权的相似度度量方式选取近邻样本集,通过局部加权偏最小二乘法构建局部模型进行LF精炼终点温度和终点硫含量的预测。基于国内某钢厂LF精炼车间实际生产数据对本预测方法进行验证,并与传统全局反向传播(back propagation,BP)建模和普通欧氏距离相似度度量策略局部建模方法进行对比。结果表明,本预测方法对Q235A/B终点温度在±5℃范围内预测命中率达92.5%,对脱硫钢种终点硫质量分数在±0.002%范围内预测命中率达90.0%,优于其他两种方法,可以为LF精炼实际生产终点控制及后续出钢工作提供指导参考。LF refining is an important process for linking converter and continuous casting.The accurate prediction of LF end-point can improve the efficiency of LF production and ensure the stability of subsequent process.In order to improve the prediction accuracy of LF end-point,an end-point prediction method for LF refining based on just-in-time learning local model was proposed.Under the just-in-time learning framework,the similarity measurement method with dual weights of feature importance and time was used to determine the sample set of nearest neighbors,and a local model was constructed by the locally weighted partial least squares method to predict the end-point temperature and sulfur content of LF refining.The prediction method was verified based on the actual production data of LF refining workshop of a domestic steel plant,and compared with the traditional global BP modeling and the local modeling method with the general Euclidean distance similarity measurement strategy.The results show that the hitting rate of the prediction method proposed in this study is 92.5% for the end-point temperature of Q235A/B within the range of ±5℃,and 90.0% for the end-point sulfur mass fraction of desulfurization steels within the range of ±0.002%,which was superior to the other two methods.It can provide guidance and reference for the end-point control of LF refining and the subsequent tapping process in actual production.

关 键 词：LF精炼 终点预测 即时学习 局部加权偏最小二乘法 局部模型 

分 类 号：TF769.2[冶金工程—钢铁冶金]