东北大学学报(自然科学版) ›› 2025, Vol. 46 ›› Issue (10): 51-58.DOI: 10.12068/j.issn.1005-3026.2025.20240051

• 材料与冶金 • 上一篇    下一篇

RH终点钢液碳质量分数的智能预测

李登辉1,2, 赵岩2, 雷洪1,2, 范佳3   

  1. 1.东北大学 材料电磁过程研究教育部重点实验室,辽宁 沈阳 110819
    2.东北大学 冶金学院,辽宁 沈阳 110819
    3.河北钢铁集团 邯钢公司,河北 邯郸 056000
  • 收稿日期:2024-03-05 出版日期:2025-10-15 发布日期:2026-01-13
  • 作者简介:李登辉(1998—),男,甘肃定西人,东北大学硕士研究生
    雷 洪(1973—),男,湖北武汉人,东北大学教授,博士生导师.
  • 基金资助:
    国家自然科学基金资助项目(52574371)

Intelligent Prediction for Endpoint Mass Fraction of Carbon in Molten Steel of RH

Deng-hui LI1,2, Yan ZHAO2, Hong LEI1,2, Jia FAN3   

  1. 1.Key Laboratory of Electromagnetic Processes of Materials,Ministry of Education,Northeastern University,Shenyang 110819,China
    2.School of Metallurgy,Northeastern University,Shenyang 110819,China
    3.Hansteel Company,HBIS Group,Handan 056000,China. Corresponding author: LEI Hong,E-mail: leihong@epm. neu. edu. cn
  • Received:2024-03-05 Online:2025-10-15 Published:2026-01-13

摘要:

准确预测RH(Ruhrstahl Heraeus)终点钢液碳质量分数能够有效地提升连铸坯质量.为实现该目标,首先采用数据挖掘方法对RH生产数据进行预处理;然后使用灰色关联分析、Spearman相关系数和随机森林袋外误差评分法筛选出与终点碳质量分数强相关的特征变量;接着用主成分分析进行降维;最后采用XGBoost模型以及粒子群优化和鲸鱼优化算法优化后的XGBoost模型预测RH钢液终点碳质量分数.研究结果表明,灰色关联分析筛选特征效果优于Spearman秩相关系数和随机森林;经过粒子群算法和鲸鱼优化算法优化后,XGBoost模型的预测命中率显著提高.鲸鱼优化算法要优于粒子群算法.当误差范围为±5×10-6和±7×10-6时,鲸鱼群优化XGBoost模型预测命中率分别达到91.26%和98.97%.

关键词: 终点钢液碳质量分数, XGBoost算法, 粒子群优化算法, 鲸鱼优化算法, 特征筛选

Abstract:

Accurate prediction of the endpoint mass fraction of carbon in the molten steel of RH (Ruhrstahl Heraeus) can effectively improve the quality of continuously cast products. In order to realize this goal, data mining was firstly applied to preprocess the RH industrial data. Then, grey correlation analysis, Spearman correlation coefficient, and random forest out-of-bag error scoring were used to select the features that had a strong correlation with the endpoint mass fraction of carbon in the molten steel. Next, the principal component analysis method was applied to reduce the dimensions. Finally, the XGBoost model, the XGBoost model optimized by the particle swarm optimization algorithm, and the XGBoost model optimized by the whale optimization algorithm were applied to predict the endpoint mass fraction of carbon in the molten steel. The results show that grey correlation analysis is better than Spearman rank correlation coefficient and random forest in analyzing the selected features. After the optimization of the particle swarm optimization algorithm and whale optimization algorithm, the XGBoost model has a greater prediction hit rate. The XGBoost model optimized by the whale optimization algorithm is better than that by the particle swarm optimization algorithm. In the case of the XGBoost model optimized by the whale optimization algorithm, the hit rate reaches 91.26% and 98.97% if the error range is within ±5×10-6, and ±7×10-6.

Key words: endpoint mass fraction of carbon in molten steel, XGBoost algorithm, particle swarm optimization algorithm, whale optimization algorithm, feature selection

中图分类号: