基于工艺理论和卷积神经网络的烧结矿转鼓指数预测  

Prediction of sinter drum index based on convolutional neural network and process theory

作　　者：刘然 张智峰 刘小杰 李欣 李宏扬 吕庆 LIU Ran;ZHANG Zhifeng;LIU Xiaojie;LI Xin;LI Hongyang;LV Qing(College of Metallurgy and Energy,North China University of Science and Technology,Tangshan 063210,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

出　　处：《钢铁研究学报》2023年第6期651-658,共8页Journal of Iron and Steel Research

基　　金：国家自然科学基金资助项目(52004096);河北省自然科学基金资助项目(E2020209208)。

摘　　要：作为评价烧结矿质量的重要指标之一,转鼓指数的高低直接影响着高炉生产的稳定与否。以某钢铁企业烧结生产数据为基础,提出了基于特征工程与图像识别技术的烧结矿转鼓指数预测方法。首先对挑选出的3类28个影响烧结矿转鼓指数的重要指标完成数据预处理;而后通过SVM-RFE算法以及交叉验证算法筛选出对目标变量影响较大的特征参数;最后用卷积神经网络对经过数据特征转化的二维特征图像进行训练,建立了基于卷积神经网络的烧结矿转鼓指数预测模型。结果表明,在误差范围为±1%的情况下该模型命中率高达93.71%。这种将数据特征转化为图像特征的处理方法有效地提高了预测能力,对未来预测式烧结技术的发展具有很好的借鉴意义。As one of the important indicators for evaluating the quality of sintered ore,the drum index directly affects the stability of blast furnace production.Based on the sintering production data of an iron and steel enterprise,a prediction method of sinter drum index based on feature engineering and image recognition technology was proposed.Firstly,the data preprocessing was completed for the selected 3 categories of 28 important indicators that affect the sinter drum index.Then,the feature parameters that had a greater impact on the target variable were screened out through the SVM-RFE algorithm and the cross-validation algorithm.Finally,the convolutional neural network was used to train the two-dimensional feature image transformed by the data features,and a prediction model of the sinter drum index based on the convolutional neural network was established.The results show that the hit rate of the model is as high as 93.71%with an error margin of±1%.This method of converting data features into image features effectively improves the prediction ability,and has a good reference for the future development of predictive sintering technology.

关 键 词：烧结 转鼓指数 卷积神经网络 大数据 特征工程 

分 类 号：TF046[冶金工程—冶金物理化学] TP183[自动化与计算机技术—控制理论与控制工程]