| 周子轩, 朱仁传, 詹可.基于机器学习的Kelvin源格林函数数值预报[J].海洋工程,2025,(2):170~178 |
| 基于机器学习的Kelvin源格林函数数值预报 |
| Numerical prediction of Kelvin source Green’s function based on machine learning |
| 投稿时间：2024-04-29 |
| DOI：10.16483/j.issn.1005-9865.2025.02.016 |
| 中文关键词: Kelvin源格林函数 远场波动项 高频振荡 多层感知机 |
| 英文关键词:Kelvin source Green’s function far-field wave component high-frequency oscillations multilayer perceptron |
| 基金项目:国家自然科学基金资助项目（U2141228） |
|
| 摘要点击次数: 12 |
| 全文下载次数: 11 |
| 中文摘要: |
| 快速、准确地计算Kelvin源格林函数及其偏导数是求解船舶移动兴波问题的基础及核心，其主要困难源于远场波动项积分核的高频振荡。鉴于机器学习理论中多层感知机（multilayer perceptron，简称MLP）的强大非线性映射能力，构建了针对Kelvin源格林函数及其偏导数的MLP数值预报模型。模型训练及测试所需的高精度数据样本均由自适应积分方法生成。为了充分利用样本信息，降低训练耗时，提出了一种基于收敛函数的创新采样策略，实现对近自由面区域的局部加密采样。数值分析结果表明，MLP模型预报精度良好，且效率显著优于自适应积分等数值方法。该模型不仅在兴波问题求解中有重要应用价值，也为机器学习在传统水动力领域的应用开辟了新的思路。 |
| 英文摘要: |
| The rapid and accurate computation of the Kelvin source Green’s function and its partial derivatives is fundamental and central to solving the ship motion wave-making problem. The primary challenge lies in the high-frequency oscillations of the far-field wave component’s integral kernel. Given the strong nonlinear mapping capability of multilayer perceptrons (MLP) in machine learning theory, an MLP-based numerical prediction model for the Kelvin source Green’s function and its partial derivatives is developed. High-precision data samples required for model training and testing are generated using an adaptive integration method. To fully exploit sample information and reduce training time, an innovative convergence-function-based sampling strategy is proposed to achieve local refinement sampling near the free surface region. Numerical analysis results indicate that the MLP model demonstrates excellent prediction accuracy and significantly outperforms numerical methods such as adaptive integration in terms of efficiency. This model not only has significant application value in solving wave-making problems but also opens new avenues for applying machine learning in traditional hydrodynamics. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
| 关闭 |
