基于支持向量回归的高炉出铁量预测方法  

PREDICTION OF BLAST FURNACE IRON OUTPUT BASED ON SUPPORT VECTOR REGRESSION

作　　者：李建生[1] 盛钢 张硕[1] Li Jiansheng;Sheng Gang;Zhang Shuo(Production and Manufacturing Department of Tangshan Iron and Steel Group Co.,Ltd.,Tangshan 063016,Hebei;Iron and steel Research Institute Co.,Ltd.,Beijing 100089)

机构地区：[1]唐山钢铁集团有限责任公司生产制造部,河北唐山063016 [2]钢铁研究总院有限公司,北京100089

出　　处：《河北冶金》2024年第10期55-58,70,共5页Hebei Metallurgy

摘　　要：高炉炼铁是钢铁生产中的重要环节。基于高炉流出的铁水量,选择指定数量的铁包进行装载和运输,能够提高生产效率,并降低整体调度线的能耗。准确预测出铁量,对于后续生产调度有着重要意义。但一方面高炉炼铁涉及大量的物理化学反应和参数变化,且炼铁过程无法从外部实时观测,难以通过直接进行机理分析实现准确的自动控制;另一方面,炼铁过程中记录的鼓风参数、焦炭比、炉渣成分等参数丰富的测量数据,可被用于数据驱动的建模分析。本文旨在通过机理模型分析理想状态下的铁水流速,并设计基于支持向量回归的机器学习模型,对高炉出铁量进行预测。对某日产8000 t铁量高炉的出铁数据进行建模分析,实验结果表明,支持向量回归模型预测出铁量的平均误差在200 t以内,且平均误差、预测标准差等指标优于其它常见的机器学习模型,表现出了数据驱动模型的准确性,能够对实际的高炉炼铁分析和建模提供指导作用,从而降低资源消耗,并提高整体钢铁生产线的生产效率。Blast furnace ironmaking is an important link in steel production.Based on the amount of iron water flowing out of the blast furnace,selecting a specified number of iron bags for loading and transportation can improve production efficiency and reduce the overall consumption of the scheduling line.Accurately predicting the amount of iron is of great significance for subsequent production scheduling.On the one hand,blast furnace ironmaking involves a large number of physical and chemical reactions and parameter changes,and the ironmaking process cannot be observed in real-time from the outside,making it difficult to achieve accurate automatic control through direct mechanism analysis;On the other hand,the rich measurement data of blast parameters,coke ratio,slag composition and other parameters recorded during the ironmaking process can be used for data-driven modeling and analysis.This article aims to analyze the ideal flow rate of molten iron through a mechanistic model and design a machine learning model based on support vector regression to predict the iron output of blast furnaces.The experimental results of modeling and analyzing the iron production data of a blast furnace with a daily output of 8000 tons show that the average error of the support vector regression model in predicting the iron production is within 200 tons,and the average error,prediction standard deviation and other indicators are better than other common machine learning models,demonstrating the accuracy of the data-driven model.It can provide guidance for actual blast furnace ironmaking analysis and modeling,thereby reducing resource consumption and improving the overall production efficiency of the steel production line.

关 键 词：高炉炼铁 出铁量 机器学习 支持向量回归 数据驱动 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]