摘要
高炉热制度是高炉生产的基本制度。对于复杂、难控制的小时级延迟的高炉系统,掌握高炉铁水Si含量的变化趋势对改善高炉炉热水平和维持炉况顺行具有重要意义。目前的铁水Si含量预测模型在建模数据样本维度、样本数量、预测准确度与可靠性等方面尚存在不足,限制了模型的工业化应用。提出了一种基于数据驱动的高炉铁水Si含量预测模型,在建模过程中,全面分析了整个高炉工序包括原燃料、工艺操作、冶炼状态和渣铁排放过程的数据。该模型共包括152个变量、9223组数据。通过特征过滤和特征选择过程确定了高炉铁水Si含量预测模型的输入特征。相较于传统单一机器学习算法,该模型采用Stacking框架集成学习方法,模型性能明显提升,并通过与其他5种机器学习算法对比,验证了所提方法的性能。经统计,该算法预测的铁水Si含量在-0.05%~0.05%误差范围内的命中率达到了90.48%。
Thermal state is a basic system of blast furnace production.For the complex and difficult-to-control blast furnace system with hourly delay,it is of great significance to grasp the trend of Si content of blast furnace hot metal in advance in order to improve the blast furnace heat level and furnace smoothness.In this paper,a data-driven prediction model of Si content of blast furnace hot metal was proposed.In the modeling process,the data of raw fuel,process operation,smelting status,and slag-iron discharge of the entire blast furnace process were comprehensively analyzed,with a total of 152 variables and 9223 sets of data.And the input features for determining the prediction model of Si content of blast furnace hot metal were identified through the feature filtering and feature selection process,which played an important role in the modeling.Compared with the traditional single machine learning algorithm,the model performance of the integrated learning method using Stacking framework was significantly improved,and the performance of the proposed method was verified by comparing it with other five machine learning algorithms.According to statistics,the hit rate of Si content of the hot metal predicted by this algorithm within the error range of -0.05% to 0.05% is 90.48%.
作者
李云涛
王士彬
张代华
LI Yuntao;WANG Shibin;ZHANG Daihua(Research Institute,Baoshan Iron&Steel Co.,Ltd.,Shanghai 201999,China;Iron-making Plant,Baoshan Iron&Steel Co.,Ltd.,Shanghai 200941,China)
出处
《宝钢技术》
2024年第6期11-16,共6页
Baosteel Technology
关键词
铁水Si含量
数据驱动
特征选择
集成学习
Si content of the hot metal
data-driven approach
feature selection
ensemble learning
