Scholarlink

<p>In the converter steelmaking process, meeting the standard requirement by the contents of several elements in the final molten steel is mandatory for the converter melting being qualified. Conventional methods such as molten steel sampling, composition testing, and result reporting are time-consuming and causing delays, thus hindering the overall smelting efficiency. To this end, the current study developed a converter sample-free steel pouring prediction system. A database was established by obtaining 783 datasets, such as the initial composition of molten iron, smelting parameters in the converter process, and C and P elements in the molten steel at the converter endpoint (CEP) during the industrial production process. The prediction of C and P elements at the CEP was realized using the backpropagation neural network (BPNN), gradient boosting decision tree (GBDT), and random forest (RF) algorithms, the latter exhibiting the best prediction accuracy. The main input features for predicting elements C and P were found to be the weight of molten iron and scrap steel, respectively, using the feature importance ranking. The RF algorithm was improved using the gray wolf optimization (GWO) algorithm, and the prediction results were adjusted using the above two most significant features. In the carbon content predictions, the correlation coefficient <i>R</i><sup>2</sup> of the improved GWO-RF algorithm increased from 85.367 to 90.327 pct, while that of element P increased from 80.464 to 86.418 pct. The total prediction time at the converter endpoint was controlled within 1 second. The algorithm was incorporated into a self-developed prediction software. Good results were achieved through on-site tests in the industrial field, improving converter smelting efficiency.</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

Real-Time Prediction of C and P Contents in BOF Steelmaking Using Machine Learning Without Sample Analysis Delay

  • 期刊论文

  • 发表日期：

    2026年6月30日

查看全文
  • Xingping Jiang,
  • Haihua Wang,
  • Yongjiang Kong,
  • Guoyong Shuai,
  • Zhongdui Tang,
  • Jinfa Ding,
  • Gang Yuan,
  • Xiaohang Liu,
  • Qiang Wang

摘要

In the converter steelmaking process, meeting the standard requirement by the contents of several elements in the final molten steel is mandatory for the converter melting being qualified. Conventional methods such as molten steel sampling, composition testing, and result reporting are time-consuming and causing delays, thus hindering the overall smelting efficiency. To this end, the current study developed a converter sample-free steel pouring prediction system. A database was established by obtaining 783 datasets, such as the initial composition of molten iron, smelting parameters in the converter process, and C and P elements in the molten steel at the converter endpoint (CEP) during the industrial production process. The prediction of C and P elements at the CEP was realized using the backpropagation neural network (BPNN), gradient boosting decision tree (GBDT), and random forest (RF) algorithms, the latter exhibiting the best prediction accuracy. The main input features for predicting elements C and P were found to be the weight of molten iron and scrap steel, respectively, using the feature importance ranking. The RF algorithm was improved using the gray wolf optimization (GWO) algorithm, and the prediction results were adjusted using the above two most significant features. In the carbon content predictions, the correlation coefficient R2 of the improved GWO-RF algorithm increased from 85.367 to 90.327 pct, while that of element P increased from 80.464 to 86.418 pct. The total prediction time at the converter endpoint was controlled within 1 second. The algorithm was incorporated into a self-developed prediction software. Good results were achieved through on-site tests in the industrial field, improving converter smelting efficiency.