基于数据聚类及特征提取的高炉操作炉型  

High-temperature furnace type operation optimization based on data clustering and feature extraction

作　　者：吴亚明 黄云 陶林和 吴志慷 蔡学斌 左海滨[2] WU Yaming;HUANG Yun;TAO Linhe;WU Zhikang;CAI Xuebin;ZUO Haibin(Xichang Steel Vanadium Co.,Ltd.,Panzhihua Iron and Steel Group,Xichang 615000,Sichuang,China;State Key Laboratory of Advanced Metallurgy,University of Science and Technology Beijing,Beijing 100083,China)

机构地区：[1]攀钢集团西昌钢钒有限公司,四川西昌615000 [2]北京科技大学绿色低碳钢铁冶金全国重点实验室,北京100083

出　　处：《钢铁》2025年第3期45-55,共11页Iron and Steel

摘　　要：合理操作炉型是高炉长寿、稳定顺行和获得良好经济技术指标的关键。基于某钢厂高炉生产数据,探究高炉炉型优化方法,为高炉操作提供科学指导。首先采用孤独森林法和箱型图法对数据进行噪声识别和处理,并基于主成分分析(PCA)进行降维,消除噪声和数据冗余,为后续聚类分析提供高质量的数据基础。随后,对比了K-means和DBSCAN 2种聚类算法的应用效果。K-means算法在聚类簇数为14时得到最佳轮廓系数,表明高炉炉型可分为14类;DBSCAN算法在邻域半径Eps和最小邻域样本数min_samples为6.25和2时表现出更低的戴维森堡丁指数,聚类效果最佳,并能有效识别任意形状的聚类,尤其适合处理高炉生产数据的复杂性和非线性特征。为了评估不同炉型的优劣,建立了基于生产指标综合评分的操作炉型评价方法,选取焦比、燃料比、产量和铁损作为关键绩效指标,并赋予不同的权重。结果表明,第4类炉型在高炉操作指标方面表现最优,可作为合理炉型的操作目标。为了实现高炉炉型优化,利用随机森林法探究了高炉操作参数与炉型之间的隐式关系,确定了影响炉型的关键特征参数,包括布料矩阵参数、透气性指数、煤气利用率和标准风速等。通过分析炉型演变过程和高炉参数变化趋势,发现炉型恶化主要与透气性下降有关,进而导致气流分布不均、煤气利用率降低和压差上升。建立了优化高炉炉型管理的新方法,可为现场人员提供有价值的数据分析和操作指导,有助于提高高炉操作水平,降低能耗和成本,实现高炉长寿、稳定顺行和高效生产。Reasonable operation of the furnace type is the key to its long life,stable operation,and sound economic and technical indicators for the blast furnace.Based on the production data of a certain steel plant's blast furnace,the blast furnace type optimization method was studied,providing scientific guidance for blast furnace operation.First,the Isolation Forest and Boxplot methods were adopted to identify and process noise in the data,and then Principal Component Analysis(PCA)was used for dimensionality reduction to eliminate noise and data redundancy,providing a high-quality data foundation for subsequent cluster analysis.Next,the application effects of two clustering algorithms,K-means and DBSCAN,were compared.The K-means algorithm achieved the best silhouette coefficient when the number of clusters was 14,indicating that the blast furnace type could be divided into 14 categories;the DBSCAN algorithm exhibited a lower Davies-Bouldin Index(DBI)when Neighborhood Radius(Eps)and Minimum Neighborhood Sample Count(min_samples)were 6.25 and 2,showing the best clustering effect and the ability to effectively identify clusters of any shape,especially suitable for handling the complexity and nonlinearity of blast furnace production data.To evaluate the advantages and disadvantages of different furnace types,an evaluation method for operating furnace types based on comprehensive production indicators was established,selecting coke ratio,fuel ratio,output,and iron loss as key performance indicators and assigning different weights.The results show that the fourth type of furnace type performs the best in terms of blast furnace operation indicators and can be used as the operating target for a reasonable furnace type.To achieve blast furnace type optimization,the implicit relationship between blast furnace operating parameters and furnace types was explored using the Random Forest method,determining the key feature parameters that affect furnace types,including burden matrix parameters,permeability index,gas utilization rate,a

关 键 词：操作炉型 聚类分析 主成分分析 DBSCAN法 K-means法 异常值识别 孤独森林法 箱型图法 

分 类 号：TF5[冶金工程—钢铁冶金]