Scholarlink

<p>Liquefaction is a phenomenon where water-saturated soil loses strength and behaves like a liquid due to ground shaking, leading to destructive consequences. This study investigates the use of ten machine learning (ML) models, namely random forest, extreme gradient boosting, gradient boosting (GB), light gradient boosting machine, decision tree, adaptive boosting, categorical boosting, k-nearest neighbor, voting regressor, and support vector machine, to predict the probability of liquefaction using both standard penetration test (SPT) and cone penetration test (CPT) data. A total of 220 SPT and 250 CPT datasets were utilized, with input parameters including earthquake magnitude, peak ground acceleration, total stress, effective stress, fine content, depth, SPT blow count (N<sub>1</sub>)<sub>60</sub>, and CPT tip resistance (q<sub>c</sub>). Model performance was evaluated using trend-measuring metrics (coeffiecient of determination (R<sup>2</sup>), Willmott's index, and A-20 index) and error-measuring metrics (root mean square error, mean absolute error, and mean bias error). The GB model exhibited the highest performance for both SPT (R<sup>2</sup> = 0.836) and CPT (R<sup>2</sup> = 0.832) datasets in the testing phase. Further analyses, including regression curve analysis, comprehensive measure analysis, objective function criteria, Williams plot, external validation, and most likely rank, consistently supported the superiority of the gradient boosting model. A sensitivity analysis using SHAP value, identified (N<sub>1</sub>)<sub>60</sub> and q<sub>c</sub> as the most influential input parameters for SPT and CPT, respectively. This study demonstrates the potential of ML models, particularly GB, for accurate liquefaction assessment using both SPT and CPT data, providing a reliable and efficient approach for geotechnical 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

Liquefaction assessment of soil based on SPT and CPT data using novel machine learning techniques: a practical solution

  • 期刊论文

  • 发表日期：

    2025年4月16日

查看全文
  • Rashid Mustafa,
  • Abhishek Prasad Singh,
  • Sufyan Ghani

摘要

Liquefaction is a phenomenon where water-saturated soil loses strength and behaves like a liquid due to ground shaking, leading to destructive consequences. This study investigates the use of ten machine learning (ML) models, namely random forest, extreme gradient boosting, gradient boosting (GB), light gradient boosting machine, decision tree, adaptive boosting, categorical boosting, k-nearest neighbor, voting regressor, and support vector machine, to predict the probability of liquefaction using both standard penetration test (SPT) and cone penetration test (CPT) data. A total of 220 SPT and 250 CPT datasets were utilized, with input parameters including earthquake magnitude, peak ground acceleration, total stress, effective stress, fine content, depth, SPT blow count (N1)60, and CPT tip resistance (qc). Model performance was evaluated using trend-measuring metrics (coeffiecient of determination (R2), Willmott's index, and A-20 index) and error-measuring metrics (root mean square error, mean absolute error, and mean bias error). The GB model exhibited the highest performance for both SPT (R2 = 0.836) and CPT (R2 = 0.832) datasets in the testing phase. Further analyses, including regression curve analysis, comprehensive measure analysis, objective function criteria, Williams plot, external validation, and most likely rank, consistently supported the superiority of the gradient boosting model. A sensitivity analysis using SHAP value, identified (N1)60 and qc as the most influential input parameters for SPT and CPT, respectively. This study demonstrates the potential of ML models, particularly GB, for accurate liquefaction assessment using both SPT and CPT data, providing a reliable and efficient approach for geotechnical engineering applications.