Scholarlink

<p>Enhancing operational stability is an effective approach to promote energy conservation and consumption reduction in blast furnace (BF) ironmaking. To address the limitations of traditional blast furnace condition evaluation methods, this study proposes a data-driven integrated model for comprehensive blast furnace condition evaluation, prediction, and root-cause tracing. Utilizing actual production data from a 4000&#xa0;m<sup>3</sup> blast furnace in a North China ironmaking plant, multi-source heterogeneous data underwent preprocessing, including data cleansing, standardization, and frequency alignment, to construct a high-quality smelting dataset. With fuel ratio, hot metal silicon content [Si], and gas utilization rate as target parameters, a comprehensive evaluation system was established through correlation analysis. Factor analysis reduced dimensionality of selected parameters, extracting 10 factors that collectively explained 75.52% of the original data variance. Based on factor scores, the Evaluation Index of Blast Furnace Status (EIBFS) was derived to accurately characterize blast furnace operational status. Time-lag effects were quantified via the Maximal Information Coefficient (MIC), revealing that lower-adjustment parameters (e.g., blast volume, coal injection) exhibit lags of 0 ~ 1&#xa0;h, while upper-adjustment parameters (e.g., burden distribution, coke quality) show lags of 0 ~ 4&#xa0;h. Prediction models employing Random Forest (RF), Support Vector Machine (SVM), Extreme Learning Machine (ELM), and Gradient Boosting (GB) algorithms were optimized by Bayesian Optimization (BO). Results demonstrate that the optimized SVM model achieved a 98.7% prediction hit rate within a 0.1 error tolerance, significantly outperforming other models. SHapley Additive exPlanations (SHAP) analysis precisely identified key anomaly drivers (e.g., oxygen flow rate, pulverized coal injection) and diagnosed instability causes. This model provides a scientific foundation for evaluating, predicting, and controlling blast furnace stability, advancing blast furnace ironmaking from a black-box process toward transparent and intelligent operation.</p> Graphical Abstract <p></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

A Comprehensive Evaluation–Prediction–Tracing Model for the Condition of Blast Furnace Based on Data

  • 期刊论文

  • 发表日期：

    2025年8月14日

查看全文
  • Wenchao Zhang,
  • Xiaobo Yu,
  • Kai Wang,
  • Xuan Ding,
  • Mingyin Kou,
  • Haifeng Li,
  • Shengli Wu,
  • Heng Zhou

摘要

Enhancing operational stability is an effective approach to promote energy conservation and consumption reduction in blast furnace (BF) ironmaking. To address the limitations of traditional blast furnace condition evaluation methods, this study proposes a data-driven integrated model for comprehensive blast furnace condition evaluation, prediction, and root-cause tracing. Utilizing actual production data from a 4000 m3 blast furnace in a North China ironmaking plant, multi-source heterogeneous data underwent preprocessing, including data cleansing, standardization, and frequency alignment, to construct a high-quality smelting dataset. With fuel ratio, hot metal silicon content [Si], and gas utilization rate as target parameters, a comprehensive evaluation system was established through correlation analysis. Factor analysis reduced dimensionality of selected parameters, extracting 10 factors that collectively explained 75.52% of the original data variance. Based on factor scores, the Evaluation Index of Blast Furnace Status (EIBFS) was derived to accurately characterize blast furnace operational status. Time-lag effects were quantified via the Maximal Information Coefficient (MIC), revealing that lower-adjustment parameters (e.g., blast volume, coal injection) exhibit lags of 0 ~ 1 h, while upper-adjustment parameters (e.g., burden distribution, coke quality) show lags of 0 ~ 4 h. Prediction models employing Random Forest (RF), Support Vector Machine (SVM), Extreme Learning Machine (ELM), and Gradient Boosting (GB) algorithms were optimized by Bayesian Optimization (BO). Results demonstrate that the optimized SVM model achieved a 98.7% prediction hit rate within a 0.1 error tolerance, significantly outperforming other models. SHapley Additive exPlanations (SHAP) analysis precisely identified key anomaly drivers (e.g., oxygen flow rate, pulverized coal injection) and diagnosed instability causes. This model provides a scientific foundation for evaluating, predicting, and controlling blast furnace stability, advancing blast furnace ironmaking from a black-box process toward transparent and intelligent operation.

Graphical Abstract