Scholarlink

<p>The accuracy of conventional beam models in predicting lateral pile response decreases as the pile diameter increases, due to their inability to capture complex lateral behavior when considering only lateral resistance. This study presents an integrated PINNs (Physics-Informed Neural Networks)-XGBoost (eXtreme Gradient Boosting) framework for predicting the deformation of offshore monopiles. The proposed framework accounts for the lateral deformation characteristics of both flexible and rigid piles by incorporating base resistance at the bottom boundary of the beam model. Furthermore, an XGBoost model, trained on an extensive dataset, reconstructs the distribution of soil springs along the pile, thereby supplying the PINNs with boundary conditions that realistically approximate soil-structure interactions for solving the governing beam equations. The effectiveness of this approach is validated against multiple centrifuge and field tests. The lateral responses of large-diameter monopiles under scoured conditions are also evaluated, demonstrating the model’s applicability in complex offshore environments. By eliminating the need for repetitive modeling, the approach enables efficient and comprehensive assessment of scour effects across varying depths and morphologies throughout the scour evolution process.</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 PINNs-XGBoost integrated framework for offshore monopile deformation prediction

  • 期刊论文

  • 发表日期：

    2026年5月7日

查看全文
  • Biao Li,
  • Wengang Qi,
  • Shunyi Wang,
  • Qingkai Song,
  • Fuping Gao

摘要

The accuracy of conventional beam models in predicting lateral pile response decreases as the pile diameter increases, due to their inability to capture complex lateral behavior when considering only lateral resistance. This study presents an integrated PINNs (Physics-Informed Neural Networks)-XGBoost (eXtreme Gradient Boosting) framework for predicting the deformation of offshore monopiles. The proposed framework accounts for the lateral deformation characteristics of both flexible and rigid piles by incorporating base resistance at the bottom boundary of the beam model. Furthermore, an XGBoost model, trained on an extensive dataset, reconstructs the distribution of soil springs along the pile, thereby supplying the PINNs with boundary conditions that realistically approximate soil-structure interactions for solving the governing beam equations. The effectiveness of this approach is validated against multiple centrifuge and field tests. The lateral responses of large-diameter monopiles under scoured conditions are also evaluated, demonstrating the model’s applicability in complex offshore environments. By eliminating the need for repetitive modeling, the approach enables efficient and comprehensive assessment of scour effects across varying depths and morphologies throughout the scour evolution process.