基于多类别生产状态的烧结矿转鼓指数预测模型  

Predictive model of sinter drum index based on multi-category production status

作　　者：张振 李欣 刘颂 李福民 刘小杰 吕庆 ZHANG Zhen;LI Xin;LIU Song;LI Fu-min;LIU Xiao-jie;LÜQing(School of Metallurgy and Energy,North China University of Science and Technology,Tangshan 063210,Hebei,China;Department of Computer Science and Technology,Tangshan College,Tangshan 063000,Hebei,China)

机构地区：[1]华北理工大学冶金与能源学院,河北唐山063210 [2]唐山学院计算机科学与技术系,河北唐山063000

出　　处：《中国冶金》2022年第1期27-35,共9页China Metallurgy

基　　金：河北省自然科学基金高端钢铁冶金联合基金项目(E2019209314);河北省教育厅科学技术研究项目资助(BJ2021099)。

摘　　要：将烧结生产大数据与机器学习算法相结合,提出了一种多类别生产状态下预测烧结矿转鼓指数的研究方法。通过数据采集、整合及预处理操作,共获得特征参数65种。以烧结终点位置(BTP)为基础,采用试验研究及可视化分析等方法将转鼓指数划分为2个类别。基于分类别转鼓指数数据集,使用特征选择算法计算了特征参数的重要排序,确定最佳特征参数组合作为模型输入参数,使用LightGBM和CatBoost算法分别建立了转鼓指数的预测模型。测试结果表明,CatBoost预测模型综合预测效果最好,与全部数据集构建的转鼓指数预测模型相比,分类别构建的非正常烧和正常烧转鼓指数预测模型的预测效果均得到一定提升,决定系数R;拟合度可达88.09%和90.69%。同时,多类别生产状态下的烧结矿转鼓指数预测模型在误差范围0.25%内命中率能够达到95%。This paper combined big data of sintering production with machine learning algorithms, and proposed a research method for predicting sinter drum index under multi-category production conditions. Through collection, integration and pre-processing operations of data, a total of 65 characteristic parameters were obtained. Based on the sintering end point(BTP), the drum index was divided into two categories used experimental research and visual analysis. Based on the classification of drum index data set, the feature selection algorithm was used to calculate the important ranking of feature parameters, and the best combination of feature parameters was determined as the model input parameters. The LightGBM and CatBoost algorithms were used to establish the prediction models of drum index respectively. The test results showed that the CatBoost prediction model had the best comprehensive prediction effect. Compared with drum index prediction model constructed by all data sets, the prediction effects of abnormal and normal drum index prediction models constructed by categories had been improved to a certain extent. The R;fit degree could reach 88.09% and 90.69%. In addition, the prediction model of sinter drum index under multi-category production state could achieve a hit rate of 95% within the error range of 0.25%.

关 键 词：铁矿石烧结 多类别生产状态 转鼓指数预测 BTP 机器学习算法 

分 类 号：TF046.4[冶金工程—冶金物理化学]