Scholarlink

<p>The blast furnace (BF) gas utilization rate (GUR) is a crucial index in BF ironmaking, and accurately predicting it is vital for optimizing energy consumption, reducing costs, and ensuring smooth BF operation. However, traditional prediction methods rely on limited input parameters under stable BF conditions, overlooking dynamic changes in input correlations during fluctuating BF conditions and thereby significantly undermining prediction accuracy. In this study, a gated recurrent unit (GRU) model based on a temporal pattern attention (TPA) mechanism for GUR prediction is proposed. After filtering out outliers from the dataset, TPA is used to dynamically evaluate and automatically assign weights to input parameter used to train the GRU networks. Compared with existing machine learning approaches, such as GRU, Least Absolute Shrinkage, and Selection Operator, combined with Long Short-Term Memory (LASSO-LSTM) and TPA combined with the the Long Short-Term Memory (TPA-LSTM) model, the proposed method achieves RMSE and MAE values of 0.0971 and 0.0515, exhibiting superior prediction performance for GUR. This study provides an effective strategy for accurate GUR prediction, supporting efficient and stable BF operation.</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

Prediction of Blast Furnace Gas Utilization Rate Using a Temporal Pattern Attention Mechanism-Enhanced Gated Recurrent Unit Network

  • 期刊论文

  • 发表日期：

    2026年5月28日

查看全文
  • Xiangdong Xing,
  • Haoxuan Chen,
  • Penghui Guo,
  • Lishuo Wang,
  • Kailun Zhang,
  • Hongmin Guo,
  • Bin Li,
  • Donghui Wei

摘要

The blast furnace (BF) gas utilization rate (GUR) is a crucial index in BF ironmaking, and accurately predicting it is vital for optimizing energy consumption, reducing costs, and ensuring smooth BF operation. However, traditional prediction methods rely on limited input parameters under stable BF conditions, overlooking dynamic changes in input correlations during fluctuating BF conditions and thereby significantly undermining prediction accuracy. In this study, a gated recurrent unit (GRU) model based on a temporal pattern attention (TPA) mechanism for GUR prediction is proposed. After filtering out outliers from the dataset, TPA is used to dynamically evaluate and automatically assign weights to input parameter used to train the GRU networks. Compared with existing machine learning approaches, such as GRU, Least Absolute Shrinkage, and Selection Operator, combined with Long Short-Term Memory (LASSO-LSTM) and TPA combined with the the Long Short-Term Memory (TPA-LSTM) model, the proposed method achieves RMSE and MAE values of 0.0971 and 0.0515, exhibiting superior prediction performance for GUR. This study provides an effective strategy for accurate GUR prediction, supporting efficient and stable BF operation.