Scholarlink

<p>Pile settlement in soft soil is a critical issue that must be carefully considered in the design and construction of deep foundations, especially under complex geotechnical conditions such as silty clay. While the finite element method (FEM) offers high accuracy in simulating pile-soil interaction, it often requires significant computational time and depends heavily on commercial software, limiting its widespread application in preliminary design stages. To address this limitation, the present study develops a machine learning model based on the XGBoost algorithm, optimized using a Genetic Algorithm (GA), to serve as a surrogate for FEM in predicting vertical displacement (UZ) of bored piles. The training dataset comprises 910 FEM simulations, based on 17 laboratory-tested soil samples representing a diverse range of geotechnical properties typically found in soft ground conditions. The GA is employed to fine-tune hyperparameters of the XGBoost model, thereby enhancing its prediction accuracy and mitigating the risk of overfitting. Evaluation results show that the model achieves a coefficient of determination R<sup>2</sup> = 0.8235, a root mean square error (RMSE) of approximately 217.97&#xa0;mm, and a mean absolute error (MAE) of around 57.64&#xa0;mm, demonstrating the model’s ability to capture nonlinear behavior and predict UZ with high reliability. Given its strong predictive performance and significantly reduced computation time compared to FEM, the GA-XGBoost model is proposed as an efficient and practical tool for supporting foundation design, particularly in preliminary assessment or when large-scale parametric simulations are required.</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

Enhancing Bored Pile Settlement Prediction in Silty Clay Using a Genetic Algorithm Optimized XGBoost Model

  • 期刊论文

  • 发表日期：

    2025年8月8日

查看全文
  • Luan Vo Nhat,
  • Tuan Nguyen Anh,
  • Hoa Tran Vu Van

摘要

Pile settlement in soft soil is a critical issue that must be carefully considered in the design and construction of deep foundations, especially under complex geotechnical conditions such as silty clay. While the finite element method (FEM) offers high accuracy in simulating pile-soil interaction, it often requires significant computational time and depends heavily on commercial software, limiting its widespread application in preliminary design stages. To address this limitation, the present study develops a machine learning model based on the XGBoost algorithm, optimized using a Genetic Algorithm (GA), to serve as a surrogate for FEM in predicting vertical displacement (UZ) of bored piles. The training dataset comprises 910 FEM simulations, based on 17 laboratory-tested soil samples representing a diverse range of geotechnical properties typically found in soft ground conditions. The GA is employed to fine-tune hyperparameters of the XGBoost model, thereby enhancing its prediction accuracy and mitigating the risk of overfitting. Evaluation results show that the model achieves a coefficient of determination R2 = 0.8235, a root mean square error (RMSE) of approximately 217.97 mm, and a mean absolute error (MAE) of around 57.64 mm, demonstrating the model’s ability to capture nonlinear behavior and predict UZ with high reliability. Given its strong predictive performance and significantly reduced computation time compared to FEM, the GA-XGBoost model is proposed as an efficient and practical tool for supporting foundation design, particularly in preliminary assessment or when large-scale parametric simulations are required.