Scholarlink

<p>Machine learning algorithms have gained prominence in ladle refining in recent years due to their superior accuracy and speed compared to traditional methods such as manual experience and numerical simulations. This study utilized a ladle water model to capture 700 oil eye area (OEA) images, which were then processed through binarization to create a comprehensive database. The dataset was divided into training and testing sets in an 8:2 ratio. Input features for the algorithms included gas flow rate, oil layer thickness, and purging plug position, while the output feature was the OEA percentage. To enhance the performance of the BPNN, DT, and XGBoost algorithms, three optimization algorithms (GA, PSO, and GWO) were employed to determine the optimal combination of hyperparameters. This approach established a predictive algorithm system for slag eye morphology in steel refining. Evaluation metrics such as MAE, MSE, MAPE, and <i>R</i><sup>2</sup> were used to assess the algorithms’ performance. Results showed that the prediction accuracy of the GA-BPNN algorithm improved from 86.54 to 90.07&#xa0;pct, the PSO-DT algorithm improved from 94.08 to 96.61&#xa0;pct, and the GWO-XGBoost algorithm improved from 94.35 to 96.10&#xa0;pct. However, comparing predicted and experimental values revealed that the GA-BPNN and PSO-DT algorithms exhibited greater error than the GWO-XGBoost algorithm for data with an SEA exceeding 30&#xa0;pct. With continuous training and Iteration, the GWO-XGBoost algorithm demonstrated superior optimization effects, real-time prediction time is reduced to less than 0.4&#xa0;seconds. Consequently, it is concluded that the GWO-XGBoost algorithm exhibits better adaptability to the slag eye database, and its integration into online prediction software and application in industrial production offers valuable guidance.</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

Enhancing Predictive Accuracy of Slag Eye Morphology in Ladle Refining: A Comparative Study of Machine Learning Algorithms

  • 期刊论文

  • 发表日期：

    2025年4月11日

查看全文
  • Xiaohang Liu,
  • Yongkang Wang,
  • Chang Liu,
  • Aida Xiao,
  • Guangqiang Li,
  • Zhu He,
  • Qiang Wang

摘要

Machine learning algorithms have gained prominence in ladle refining in recent years due to their superior accuracy and speed compared to traditional methods such as manual experience and numerical simulations. This study utilized a ladle water model to capture 700 oil eye area (OEA) images, which were then processed through binarization to create a comprehensive database. The dataset was divided into training and testing sets in an 8:2 ratio. Input features for the algorithms included gas flow rate, oil layer thickness, and purging plug position, while the output feature was the OEA percentage. To enhance the performance of the BPNN, DT, and XGBoost algorithms, three optimization algorithms (GA, PSO, and GWO) were employed to determine the optimal combination of hyperparameters. This approach established a predictive algorithm system for slag eye morphology in steel refining. Evaluation metrics such as MAE, MSE, MAPE, and R2 were used to assess the algorithms’ performance. Results showed that the prediction accuracy of the GA-BPNN algorithm improved from 86.54 to 90.07 pct, the PSO-DT algorithm improved from 94.08 to 96.61 pct, and the GWO-XGBoost algorithm improved from 94.35 to 96.10 pct. However, comparing predicted and experimental values revealed that the GA-BPNN and PSO-DT algorithms exhibited greater error than the GWO-XGBoost algorithm for data with an SEA exceeding 30 pct. With continuous training and Iteration, the GWO-XGBoost algorithm demonstrated superior optimization effects, real-time prediction time is reduced to less than 0.4 seconds. Consequently, it is concluded that the GWO-XGBoost algorithm exhibits better adaptability to the slag eye database, and its integration into online prediction software and application in industrial production offers valuable guidance.