Scholarlink

<p>In blast furnace ironmaking, the silicon content of molten iron is widely used as an indirect indicator of furnace temperature. It is therefore employed to predict and control furnace temperature, so as to ensure stable and smooth blast furnace operation. In this paper, by collecting the actual data of a blast furnace site, after data processing, the key parameters are selected as input parameters, and the silicon content of molten iron is used as the output parameter. A prediction model of molten iron silicon content in the blast furnace is then constructed by combining random forest (RF) and long short-term memory (LSTM) networks, and SHAP analysis is performed on the model outputs to identify the most influential features. The outputs of the prediction model are further combined with the selected key control parameters to construct a decision model, thereby forming a closed-loop “prediction–decision–prediction”control framework. The results show that the RF-LSTM model achieves higher prediction accuracy than a single RF model or LSTM model. The decision model stabilizes the silicon content in molten iron within 0.30% ~ 0.41% and effectively reduces the fluctuation range of the silicon content in molten iron.</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

Intelligent Prediction and Decision Model of Blast Furnace Temperature Based on Machine Learning and Process Control

  • 期刊论文

  • 发表日期：

    2026年2月5日

查看全文
  • Liu Xiaojie,
  • Su Xinxin,
  • Zhang Yujie,
  • Li Hongwei,
  • Li Jianpeng,
  • Duan Yifan,
  • Li Yanjiang

摘要

In blast furnace ironmaking, the silicon content of molten iron is widely used as an indirect indicator of furnace temperature. It is therefore employed to predict and control furnace temperature, so as to ensure stable and smooth blast furnace operation. In this paper, by collecting the actual data of a blast furnace site, after data processing, the key parameters are selected as input parameters, and the silicon content of molten iron is used as the output parameter. A prediction model of molten iron silicon content in the blast furnace is then constructed by combining random forest (RF) and long short-term memory (LSTM) networks, and SHAP analysis is performed on the model outputs to identify the most influential features. The outputs of the prediction model are further combined with the selected key control parameters to construct a decision model, thereby forming a closed-loop “prediction–decision–prediction”control framework. The results show that the RF-LSTM model achieves higher prediction accuracy than a single RF model or LSTM model. The decision model stabilizes the silicon content in molten iron within 0.30% ~ 0.41% and effectively reduces the fluctuation range of the silicon content in molten iron.

Graphical Abstract