Scholarlink

<p>The prediction of uniaxial compressive strength (UCS) is crucial for assessing the classification of rock engineering and evaluating rock mechanical properties. Despite various optimization algorithms integrated with support vector regression (SVR) models, the optimal parameter combinations and prediction accuracy of these models are unclear. To address this, four association algorithms were employed to obtain the best input parameters for the SVR model, and nine optimization algorithms were applied for hyperparameter optimization of the UCS prediction model. A comprehensive evaluation of the predicted results was conducted using four evaluation indicators, score analysis, uncertainty analysis, and the Wilcoxon Test with a dataset of 87 rock samples utilized. The results indicated that Is(50) and Brazilian tensile strength were the two parameters most strongly correlated with the UCS and were regarded as input parameters for prediction. The SVR-based ensemble model effectively predicted the UCS. The SVR model based on the Grasshopper optimization algorithm exhibited the best predictive performance, achieving the highest composite score of 32. These findings can influence further studies on optimizing the UCS prediction models and deepen the understanding of algorithm selection, parameter settings, and evaluation measures.</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 Model for Uniaxial Compressive Strength of Rocks Based on Dual Optimization of Input Parameters and Hyperparameters

  • 期刊论文

  • 发表日期：

    2026年1月3日

查看全文
  • Tao Wen,
  • Chengyuan Lin,
  • Junrong Zhang,
  • Dexin Huang,
  • Ningsheng Chen

摘要

The prediction of uniaxial compressive strength (UCS) is crucial for assessing the classification of rock engineering and evaluating rock mechanical properties. Despite various optimization algorithms integrated with support vector regression (SVR) models, the optimal parameter combinations and prediction accuracy of these models are unclear. To address this, four association algorithms were employed to obtain the best input parameters for the SVR model, and nine optimization algorithms were applied for hyperparameter optimization of the UCS prediction model. A comprehensive evaluation of the predicted results was conducted using four evaluation indicators, score analysis, uncertainty analysis, and the Wilcoxon Test with a dataset of 87 rock samples utilized. The results indicated that Is(50) and Brazilian tensile strength were the two parameters most strongly correlated with the UCS and were regarded as input parameters for prediction. The SVR-based ensemble model effectively predicted the UCS. The SVR model based on the Grasshopper optimization algorithm exhibited the best predictive performance, achieving the highest composite score of 32. These findings can influence further studies on optimizing the UCS prediction models and deepen the understanding of algorithm selection, parameter settings, and evaluation measures.