Scholarlink

<p>An improved model (MF–GWO–RF) combines the modified formula (MF) with the gray wolf optimization (GWO) and random forest (RF) algorithms to achieve precise prediction of the slag eye area during the ladle refining process. Seven hundred oil eye photographs of the ladle water model water–oil simulation experiment were taken by a high-definition camera. The oil eye area was obtained by binarization, and the training database was established according to a training-to-validation ratio of 8:2. Taking the bottom-blown gas flow rate, the oil&#xa0;layer thickness, and the purging plug position as input variables to predict the oil eye area, the prediction evaluation indices of the gradient boosting decision tree (GBDT), RF, and the deep neural network (DNN) were compared; it was found that the correlation coefficients of GBDT, RF, and DNN were 94.065%, 98.485%, and 91.286%,&#xa0;respectively, while that of the proposed MF–GWO–RF model reached 99.399%. In 140 testing sets, the MF–GWO–RF model had two testing data relative errors of about 20% and 40%, respectively, while those of other testing groups were below 15%. This proved the MF–GWO–RF model’s superiority over the other models in predicting the oil eye area data. The GWO algorithm is utilized to perform global optimization of the RF model, thereby overcoming the limitation of the RF algorithm that relies on grid search for hyperparameter tuning.</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

Enhanced MF–GWO–RF algorithm for accurate slag eye area prediction in ladle refining

  • 期刊论文

  • 发表日期：

    2026年7月22日

查看全文
  • Xiao-Hang Liu,
  • Chang Liu,
  • Ai-Da Xiao,
  • Wen Yan,
  • Guang-Qiang Li,
  • Zhu He,
  • Qiang Wang

摘要

An improved model (MF–GWO–RF) combines the modified formula (MF) with the gray wolf optimization (GWO) and random forest (RF) algorithms to achieve precise prediction of the slag eye area during the ladle refining process. Seven hundred oil eye photographs of the ladle water model water–oil simulation experiment were taken by a high-definition camera. The oil eye area was obtained by binarization, and the training database was established according to a training-to-validation ratio of 8:2. Taking the bottom-blown gas flow rate, the oil layer thickness, and the purging plug position as input variables to predict the oil eye area, the prediction evaluation indices of the gradient boosting decision tree (GBDT), RF, and the deep neural network (DNN) were compared; it was found that the correlation coefficients of GBDT, RF, and DNN were 94.065%, 98.485%, and 91.286%, respectively, while that of the proposed MF–GWO–RF model reached 99.399%. In 140 testing sets, the MF–GWO–RF model had two testing data relative errors of about 20% and 40%, respectively, while those of other testing groups were below 15%. This proved the MF–GWO–RF model’s superiority over the other models in predicting the oil eye area data. The GWO algorithm is utilized to perform global optimization of the RF model, thereby overcoming the limitation of the RF algorithm that relies on grid search for hyperparameter tuning.