基于RFE-DNN的烧结矿性能预警模型  

Early warning model of sinter performance based on RFE-DNN

作　　者：李福民 董钰泽 刘颂 刘小杰 米舰君 吕庆 Li Fumin;Dong Yuze;Liu Song;Liu Xiaojie;Mi Jianjun;Lv Qing(North China University of Science and Technology;Tangshan University;Iron Making Division,Tangshan Iron and Steel Group Co.,Ltd.)

机构地区：[1]华北理工大学冶金与能源学院 [2]唐山学院人工智能学院 [3]唐山钢铁集团责任有限公司炼铁事业部

出　　处：《冶金能源》2023年第5期3-7,16,共6页Energy For Metallurgical Industry

基　　金：国家自然科学基金青年基金(52004096);河北省教育厅科学技术研究资助项目(BJ2021099);唐山市应用基础研究科学计划项目(21130233C)。

摘　　要：针对某钢厂铁前数据库中烧结物料的预警空值与预警模型不完善问题,提出了一种烧结矿性能预警模型。将传统烧结工艺理论与大数据技术相结合,对原厂烧结生产数据进行预处理并搭建相应的烧结数据仓库,运用RFE(递归特征消除)对生产参数进行特征选择、重要性排序与相关性分析,然后运用DNN算法构建烧结矿化学成分与质量指标的预测模型,预测V_(2)O_(5)、CaO/SiO_(2)、TFe和FeO的R^(2)分别达到0.9658、0.8247、0.8462和0.8711,预测筛分指数和转鼓指数的R^(2)分别达到0.899和0.875,满足预测精度需求,并将预测结果结合预警区间对烧结矿性能进行预警。An early warning model of sinter performance was proposed to solve the problems of the blank value and the imperfection of the early warning model of sinter material in the iron front database of a steel mill.Combining the traditional sintering process theory with big data technology,pre-processing the original sintering production data and building the corresponding sintering data warehouse,using RFE(recursive feature elimination)to carry out feature selection,importance ranking and correlation analysis of production parameters,and then using DNN algorithm to build the prediction model of sinter chemical composition and quality index.The R^(2)of V_(2)O_(5),CaO/S_(i)O_(2),TFe and FeO were predicted to reach 0.9658,0.8247,0.8462 and 0.8711,respectively,and the R^(2)of screen index and drum index were predicted to reach 0.899 and 0.875,respectively,meeting the requirement of prediction accuracy,and the prediction results were combined with the warning interval for early warning of sinter performance.

关 键 词：烧结矿 预警区间 数据预处理 数据仓库 预警模型 

分 类 号：TF046.4[冶金工程—冶金物理化学]