冶金机理与贝叶斯优化XGBoost融合的VD炉精炼终点钢液温度预测  

Predicting Endpoint Temperature of Molten Steel in VD Furnace Refining Process Using Metallurgical Mechanism and Bayesian Optimization XGBoost

作　　者：徐吉 信自成 兰模 林文辉 张波 刘青[1] XU Ji;XIN Zicheng;LAN Mo;LIN Wenhui;ZHANG Bo;LIU Qing(School of State Key Laboratory of Advanced Metallurgy,University of Science and Technology Beijing,Beijing 100083,China;Jiangsu Jinheng Information Technology Co.,Ltd.,Nanjing 210031,China;Lyuliang Jianlong Industrial Co.,Ltd.,Lyuliang 032100,China)

机构地区：[1]北京科技大学绿色低碳钢铁冶金全国重点实验室,北京100083 [2]江苏金恒信息科技股份有限公司,江苏南京210031 [3]吕梁建龙实业有限公司,山西吕梁032100

出　　处：《工程科学与技术》2024年第6期63-72,共10页Advanced Engineering Sciences

基　　金：国家自然科学基金项目(52374321,51974023);建龙–北科大青年科技创新基金项目(20231235);厂协项目(20230922)。

摘　　要：在炼钢生产过程中,真空脱气精炼(VD)炉是生产高品质钢的重要设备之一,其精炼终点温度对钢液质量、生产效率和连铸顺行具有重要影响。为了实现对VD炉精炼终点钢液温度的精准控制,本文采用冶金机理和贝叶斯优化极端梯度提升(metallurgical mechanism–Bayesian optimization–extreme gradient boosting,MM–BO–XGBoost)相结合的方法建立钢液温度预测模型。首先,基于VD炉冶金机理解析,确定影响精炼终点钢液温度的主要因素;其次,使用3σ原则对实际生产数据进行预处理,剔除异常值,并采用皮尔逊相关性分析剔除对钢液温度影响较小的因素,从而确定模型的输入变量;再次,将冶金机理与XGBoost模型进行融合,对输入变量的初始特征重要性进行部分放大;最后,针对XGBoost模型的超参数寻优问题,采用贝叶斯优化(BO)对其进行超参数寻优,由此构建了MM–BO–XGBoost模型。在模型仿真过程中,对本文模型同时使用网格搜索和随机搜索进行超参数寻优,旨在对比和验证BO寻优的效果;此外,使用本文提供的数据对已有的冶金机理模型、多元线性回归模型和反向传播神经网络模型进行仿真,并与MM–BO–XGBoost模型进行性能对比。结果表明:本文提出的MM–BO–XGBoost模型的超参数优化效果最好;本文模型的预测VD炉终点钢液温度在±10℃和±15℃误差范围内的命中率分别为87.81%和96.42%,均高于其他对比模型,综合性能最优。本文构建的VD炉钢液精炼终点温度预测模型,对实现钢液温度精准控制、降低生产成本和提高VD炉精炼效率具有重要的现实意义。Objective The vacuum degassing(VD)furnace is an essential piece of equipment for producing high-quality steel.Unlike other furnaces,the VD furnace lacks a heating function,leading to a significant drop in the temperature of molten steel during the refining process.If the refining endpoint temperature of molten steel is excessively high,it results in energy waste and can even disrupt the continuous casting process.If the endpoint temperature is too low,the molten steel must be reheated in the ladle furnace(LF),severely impacting production efficiency.Therefore,the refining endpoint temperature significantly influences the quality of molten steel,production efficiency,and continuous casting.In recent years,adopting new technologies such as machine learning and big data has become crucial for transforming and upgrading steelmaking plants,facilitating the shift towards intelligent manufacturing.This study proposes a method that combines metallurgical mechanism analysis,data analysis,and machine learning to develop a temperature prediction model for VD furnace refining.Methods Initially,the key factors influencing the endpoint temperature of molten steel are identified through metallurgical mechanism analysis of the VD furnace.Then,production data from the steelmaking plant are processed following operational procedures to eliminate missing and abnor-mal data.The 3σprinciple is applied to preprocess the actual production data,removing outliers to improve data quality and ensure the accuracy of subsequent model training and prediction.Pearson correlation analysis is then employed to discard factors with a minor impact on the molten steel temperature,helping to determine the input variables for the prediction model.The dataset is randomly shuffled,with 80%selected as the training set and the remaining 20%as the testing set.Following this,based on metallurgical mechanism(MM)analysis,the initial feature import-ance in the input variables of the extreme gradient boosting(XGBoost)model is partially amplified.Finally,the mo

关 键 词：VD炉精炼 钢液温度预测 机理分析 MM–BO–XGBoost模型 

分 类 号：TF769[冶金工程—钢铁冶金]