Scholarlink

<p>Accurate prediction of drilling operation time is essential for optimizing mining efficiency and scheduling; however, existing methods often exhibit limited accuracy, weak interpretability, and poor adaptability to complex geological conditions. Therefore, this study proposes a hybrid prediction framework that integrates stepwise regression and a CNN-BiLSTM model for drilling time prediction. First, key influencing factors are identified through stepwise regression and correlation analysis, which decreases redundancy and improves model efficiency. Then, a CNN-BiLSTM model is constructed to capture both spatial feature interactions and temporal dependencies in drilling data. In addition, the KernelSHAP method is employed to interpret the model by quantifying the contribution and sensitivity of each input parameter, enhancing its engineering interpretability. Experimental results based on 2291 datasets demonstrate that the proposed model achieves high prediction accuracy (R² = 0.9839, MAPE = 6.64%), outperforming conventional models. The results also indicate that drilling diameter, rock strength, and drilling power are the dominant factors affecting drilling time. This study improves prediction performance and provides interpretable insights for parameter optimization and intelligent scheduling in mining operations, providing a practical tool for real-world engineering applications.</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

A hybrid CNN-BiLSTM model with squeeze-and-excitation attention and GSABO optimization for rock drilling time prediction in underground mines

  • 期刊论文

  • 发表日期：

    2026年6月3日

查看全文
  • Ding Liu,
  • Ning Li,
  • Qizhou Wang

摘要

Accurate prediction of drilling operation time is essential for optimizing mining efficiency and scheduling; however, existing methods often exhibit limited accuracy, weak interpretability, and poor adaptability to complex geological conditions. Therefore, this study proposes a hybrid prediction framework that integrates stepwise regression and a CNN-BiLSTM model for drilling time prediction. First, key influencing factors are identified through stepwise regression and correlation analysis, which decreases redundancy and improves model efficiency. Then, a CNN-BiLSTM model is constructed to capture both spatial feature interactions and temporal dependencies in drilling data. In addition, the KernelSHAP method is employed to interpret the model by quantifying the contribution and sensitivity of each input parameter, enhancing its engineering interpretability. Experimental results based on 2291 datasets demonstrate that the proposed model achieves high prediction accuracy (R² = 0.9839, MAPE = 6.64%), outperforming conventional models. The results also indicate that drilling diameter, rock strength, and drilling power are the dominant factors affecting drilling time. This study improves prediction performance and provides interpretable insights for parameter optimization and intelligent scheduling in mining operations, providing a practical tool for real-world engineering applications.