Scholarlink

<p>The pursuit of high-quality steel production necessitates precise control of operational parameters, particularly in the pretreatment of hot metal (PHM), where traditional methods may lead to inefficiencies. Consequently, the hierarchical fusion learning architecture (HFLA), an innovative system leveraging statistical theory, machine learning, and intelligent optimization techniques, was presented for accurate sulfur content prediction in PHM. By employing a strategy-driven fusion approach, HFLA enhances feature extraction via stacked kernels, incorporating Tikhonov regularization to guide the meta-learner predictions. Validation results from steel mill production demonstrate an impressive 4.3% increase in the coefficient of determination (<i>R</i><sup>2</sup>) compared to existing models, showcasing HFLA’s superiority. Furthermore, interpretability analysis reveals new insights into PHM, facilitating the development of a collaborative optimization system for injection flow and desulfurizer dosage. Implementation tests indicate this system yields approximately 10% in cost savings, positioning HFLA as a leading solution among current state-of-the-art approaches.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Hierarchical fusion learning architecture for desulfurization endpoint prediction and process guidance

  • 期刊论文

  • 发表日期：

    2026年7月8日

查看全文
  • Yu-Duo Zhao,
  • Si-Wei Wu,
  • Guang-Ming Cao,
  • Guo-Dong Wang

摘要

The pursuit of high-quality steel production necessitates precise control of operational parameters, particularly in the pretreatment of hot metal (PHM), where traditional methods may lead to inefficiencies. Consequently, the hierarchical fusion learning architecture (HFLA), an innovative system leveraging statistical theory, machine learning, and intelligent optimization techniques, was presented for accurate sulfur content prediction in PHM. By employing a strategy-driven fusion approach, HFLA enhances feature extraction via stacked kernels, incorporating Tikhonov regularization to guide the meta-learner predictions. Validation results from steel mill production demonstrate an impressive 4.3% increase in the coefficient of determination (R2) compared to existing models, showcasing HFLA’s superiority. Furthermore, interpretability analysis reveals new insights into PHM, facilitating the development of a collaborative optimization system for injection flow and desulfurizer dosage. Implementation tests indicate this system yields approximately 10% in cost savings, positioning HFLA as a leading solution among current state-of-the-art approaches.