Scholarlink

<p>The accurate estimation of the joint roughness coefficient (JRC) is of great significance for the assessment of the mechanical properties of rock mass. However, it can be unreliable to evaluate JRC with a single statistical parameter due to the complexity of the joint surface morphology. Therefore, eight statistical parameters are used to characterize joint roughness in this paper. To predict JRC, a new deep learning prediction model (SEC) is developed by combining the Squeeze-and-Excitation network (SE-Net) and a convolutional neural network (CNN). The new model can automatically evaluate the importance of statistical parameters, overcoming the limitations of manual feature selection. The coefficient of determination (<i>R</i><sup><i>2</i></sup>), root mean square error (<i>MSE</i>), mean absolute error (<i>MAE</i>), mean absolute percentage error (<i>MAPE</i>), and performance index (<i>PI</i>) are used as evaluation metrics for the model. The JRC prediction results of the SEC model are compared with three machine learning (ML) models and traditional empirical regression methods. Experimental results demonstrate that the SEC model exhibits strong suitability for small-sample data, with JRC predictions closer to actual values, indicating higher generalization capability and accuracy. The new model provides a novel approach for applying artificial intelligence to the JRC prediction.</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 rock joint roughness coefficient determination method incorporating the SE-Net attention mechanism and CNN

  • 期刊论文

  • 发表日期：

    2025年7月9日

查看全文
  • Mingzhe Zhou,
  • Haiying Fu,
  • Yanyan Zhao,
  • Zhihao Wang,
  • Jihuan Chen

摘要

The accurate estimation of the joint roughness coefficient (JRC) is of great significance for the assessment of the mechanical properties of rock mass. However, it can be unreliable to evaluate JRC with a single statistical parameter due to the complexity of the joint surface morphology. Therefore, eight statistical parameters are used to characterize joint roughness in this paper. To predict JRC, a new deep learning prediction model (SEC) is developed by combining the Squeeze-and-Excitation network (SE-Net) and a convolutional neural network (CNN). The new model can automatically evaluate the importance of statistical parameters, overcoming the limitations of manual feature selection. The coefficient of determination (R2), root mean square error (MSE), mean absolute error (MAE), mean absolute percentage error (MAPE), and performance index (PI) are used as evaluation metrics for the model. The JRC prediction results of the SEC model are compared with three machine learning (ML) models and traditional empirical regression methods. Experimental results demonstrate that the SEC model exhibits strong suitability for small-sample data, with JRC predictions closer to actual values, indicating higher generalization capability and accuracy. The new model provides a novel approach for applying artificial intelligence to the JRC prediction.