Scholarlink

<p>To enhance the accuracy of end-point control in converter steelmaking and reduce production costs, this study addresses the limitations of traditional prediction methods, due to data noise, nonlinear coupling, and lagging regulation, by proposing an intelligent prediction model integrating data preprocessing, end-point prediction, and reverse parameter optimization. The interquartile range (IQR) method and the empirical mode decomposition (EMD) algorithm are employed to preprocess data, eliminating outliers and noise. Based on random forest feature importance analysis, 14 key parameters, including oxygen blowing volume and scrap steel addition, are screened as input values. An empirical mode decomposition–support vector regression (EMD-SVR) end-point prediction model is constructed, achieving a hit rate of 83% for end-point C content within an error range of ± 0.03, and 89% for end-point temperature within ± 15ºC, which represents a significant improvement compared to the 60% and 66% hit rates of the model without denoising. Combined with a genetic algorithm (GA), a parameter optimization model aimed at cost minimization is established. Under constraints of end-point composition, temperature, material balance, and heat balance, process parameters are reversely optimized. Compared with conventional steelmaking processes, the average cost per ton of steel is reduced by 31 CNY for heat guided by the optimized model.</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

Converter End-Point Prediction Based on EMD-SVR and Reverse Optimization of Low-Cost Steelmaking Parameters

  • 期刊论文

  • 发表日期：

    2026年3月25日

查看全文
  • Yongkun Yang,
  • Shuan Wang,
  • Ji Zhou,
  • Yu Zhao,
  • Baorong Wang,
  • Yusai Ma,
  • Zhengjiao Zhao,
  • Xiaoming Li,
  • Naihui Yang,
  • Dongping Zhan

摘要

To enhance the accuracy of end-point control in converter steelmaking and reduce production costs, this study addresses the limitations of traditional prediction methods, due to data noise, nonlinear coupling, and lagging regulation, by proposing an intelligent prediction model integrating data preprocessing, end-point prediction, and reverse parameter optimization. The interquartile range (IQR) method and the empirical mode decomposition (EMD) algorithm are employed to preprocess data, eliminating outliers and noise. Based on random forest feature importance analysis, 14 key parameters, including oxygen blowing volume and scrap steel addition, are screened as input values. An empirical mode decomposition–support vector regression (EMD-SVR) end-point prediction model is constructed, achieving a hit rate of 83% for end-point C content within an error range of ± 0.03, and 89% for end-point temperature within ± 15ºC, which represents a significant improvement compared to the 60% and 66% hit rates of the model without denoising. Combined with a genetic algorithm (GA), a parameter optimization model aimed at cost minimization is established. Under constraints of end-point composition, temperature, material balance, and heat balance, process parameters are reversely optimized. Compared with conventional steelmaking processes, the average cost per ton of steel is reduced by 31 CNY for heat guided by the optimized model.