Scholarlink

<p>This study presents a hybrid predictive framework that integrates the eXtreme Deep Factorization Machine (xDeepFM) with the Enzyme Action Optimizer (EAO) to study the capacity of offshore pile foundations subjected to combined vertical-horizontal-moment (V-H-M) loading. The model is trained using a high-fidelity dataset consisting of 3,328 cases generated by Finite Element Limit Analysis (FELA), incorporating key dimensionless parameters such as the pile embedment ratio (<i>L/D</i>), soil heterogeneity index (<i>κ</i>), strength anisotropy ratio (<i>r</i><sub><i>e</i></sub>), normalized vertical load ratio (<i>V/V₀</i>), and load inclination angle (<i>β</i>). The proposed framework achieves excellent predictive performance (R<sup>2</sup> &gt; 0.99) on both the training and testing sets and effectively captures complex, nonlinear, and interdependent patterns within the data. Model interpretability, assessed using SHapley Additive exPlanations (SHAP) values and correlation analysis, reveals strong alignment with fundamental geotechnical principles, thereby enhancing transparency and trust in the predictions. Overall, the xDeepFM-EAO framework offers a robust, accurate, and computationally efficient tool for the design and analysis of offshore foundations subjected to multidirectional loading.</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

Bio-inspired deep learning for predicting offshore pile capacity under V-H-M loads: an xDeepFM-EAO framework

  • 期刊论文

  • 发表日期：

    2025年10月22日

查看全文
  • Katavut Vichai,
  • Jim Shiau,
  • Duy Tan Tran,
  • Mohammad Khajehzadeh,
  • Suraparb Keawsawasvong,
  • Pitthaya Jamsawang

摘要

This study presents a hybrid predictive framework that integrates the eXtreme Deep Factorization Machine (xDeepFM) with the Enzyme Action Optimizer (EAO) to study the capacity of offshore pile foundations subjected to combined vertical-horizontal-moment (V-H-M) loading. The model is trained using a high-fidelity dataset consisting of 3,328 cases generated by Finite Element Limit Analysis (FELA), incorporating key dimensionless parameters such as the pile embedment ratio (L/D), soil heterogeneity index (κ), strength anisotropy ratio (re), normalized vertical load ratio (V/V₀), and load inclination angle (β). The proposed framework achieves excellent predictive performance (R2 > 0.99) on both the training and testing sets and effectively captures complex, nonlinear, and interdependent patterns within the data. Model interpretability, assessed using SHapley Additive exPlanations (SHAP) values and correlation analysis, reveals strong alignment with fundamental geotechnical principles, thereby enhancing transparency and trust in the predictions. Overall, the xDeepFM-EAO framework offers a robust, accurate, and computationally efficient tool for the design and analysis of offshore foundations subjected to multidirectional loading.