Assisted by computational mechanics methods and deep learning, this paper proposes a data-driven method for predicting random acoustic radiation and localizing underwater source in a shallow sea. A combined method of the pseudo excitation method (PEM), the finite element method (FEM), the virtual mass method (VM), and the image method-based boundary element method (I-BEM) is developed to predict the random acoustic radiation influenced by the random vibration of the fluid-structure interaction and the sound reflection of the sea surface and the seabed. The ConvNeXt network is trained on a huge number of acoustic field data generated by PEM/FEM-VM/I-BEM to extract power spectral density (PSD) features of sound pressure at sample field points. The trained models are ultimately employed to predict random acoustic radiation and localize underwater source. The simulation results demonstrate that the proposed method is effective and universal for random sound radiation prediction and source localization in the shallow sea.

中文翻译：

---

基于ConvNeXt网络的浅海贝壳结构随机声辐射预测与震源定位


在计算力学方法和深度学习的辅助下，本文提出了一种数据驱动的方法，用于预测随机声辐射和定位浅海中的水下源。提出了一种伪激励法（PEM）、有限元法（FEM）、虚拟质量法（VM）和基于图像法的边界元法（I-BEM）的组合方法来预测随机声辐射受流固相互作用的随机振动以及海面和海底声反射的影响。 ConvNeXt 网络在 PEM/FEM-VM/I-BEM 生成的大量声场数据上进行训练，以提取样本场点声压的功率谱密度（PSD）特征。经过训练的模型最终用于预测随机声辐射并定位水下源。仿真结果表明，该方法对于浅海随机声辐射预测和声源定位是有效且通用的。