基于经验正交函数和贝叶斯神经网络的水下声场预报研究

蒋方冰; 吴金荣; 侯倩男; 张祚祥; 莫亚枭

doi:10.11990/jheu.202506011

Chinese

您当前的位置：

首页 >

文章列表页 >

基于经验正交函数和贝叶斯神经网络的水下声场预报研究

论文 | 更新时间：2026-03-31

- 基于经验正交函数和贝叶斯神经网络的水下声场预报研究
- Underwater sound field prediction based on empirical orthogonal function and Bayesian neural network
- 哈尔滨工程大学学报 2025年46卷第8期页码：1508-1515
- 作者机构：
  
  1.中国科学院声学研究所, 北京 100190
  2.中国科学院水声环境特性重点实验室, 北京 100190
  3.中国科学院大学, 北京 100049
- 作者简介：
  
  [ "蒋方冰, 女, 助理工程师" ]
  [ "吴金荣, 男, 研究员，博士生导师" ]
- 基金信息：
  
  国家自然科学基金项目(11974375)
- DOI：10.11990/jheu.202506011
  中图分类号： O427.1
- 收稿：2025-06-06，
  
  网络首发：2025-06-24，
  
  纸质出版：2025-08-05
- 稿件说明：
移动端阅览
蒋方冰, 吴金荣, 侯倩男, 等. 基于经验正交函数和贝叶斯神经网络的水下声场预报研究[J]. 哈尔滨工程大学学报, 2025,46(8):1508-1515.

Fangbing JIANG, Jinrong WU, Qiannan HOU, et al. Underwater sound field prediction based on empirical orthogonal function and Bayesian neural network[J]. Journal of Harbin Engineering University, 2025, 46(8): 1508-1515.
蒋方冰, 吴金荣, 侯倩男, 等. 基于经验正交函数和贝叶斯神经网络的水下声场预报研究[J]. 哈尔滨工程大学学报, 2025,46(8):1508-1515. DOI： 10.11990/jheu.202506011.

Fangbing JIANG, Jinrong WU, Qiannan HOU, et al. Underwater sound field prediction based on empirical orthogonal function and Bayesian neural network[J]. Journal of Harbin Engineering University, 2025, 46(8): 1508-1515. DOI： 10.11990/jheu.202506011.

摘要

在水下声场预报中，数据驱动模型的预报精度主要取决于训练样本数对样本空间的覆盖程度。针对现有方法多局限于单一水文环境、且水文样本数量不足导致精度下降的问题，本文提出一种基于经验正交函数和贝叶斯神经网络的水下声场预报方法。利用经验正交函数有效降低声速剖面输入维度，并通过其系数组合生成覆盖多样化水文环境的样本集；进而借助具有强泛化能力的贝叶斯神经网络在部分数据空间内学习有效特征，预报变化水文条件下的声传播损失，并给出置信区间。结果表明: 相较于传统神经网络，该方法在训练集范围内的预报误差更小，对未知数据的适应能力更强，且通过概率建模可实现端到端的不确定性量化

提升了数据驱动模型在复杂水文条件下的鲁棒性与可靠性。

Abstract

The accuracy of data-driven models for forecasting underwater acoustic fields primarily depends on the sample space covered by the training sample. To address the limitations of existing methods

which are often confined to single sound speed profiles (SSPs) and suffer from accuracy loss due to insufficient SSP samples

this study proposes a method employing an empirical orthogonal function and Bayesian neural network (EOF-BNN). First

the input dimension of the SSPs is effectively reduced using the EOF

generating diverse SSP samples via coefficient combinations. The method then employs a BNN with strong generalization ability to learn effective features from limited data

forecasting acoustic transmission loss under varying SSP conditions while providing confidence intervals. Compared with the traditional neural network

this method has smaller prediction error in the training set range

better adaptability to unknown data

and end-to-end uncertainty quantification through probabilistic modeling

which improves the robustness and reliability of data-driven models under complex hydrological conditions.

关键词

Keywords

references

李启虎. 水声学研究进展[J]. 声学学报, 2001, 26(4): 295-301.

LI Qihu. Advances of research work in underwater acoustics[J]. Acta acustica, 2001, 26(4): 295-301.

MALLIK W, JAIMAN R K. Predicting transmission loss in underwater acoustics using convolutional recurrent autoencoder network[J]. The journal of the acoustical society of America, 2022, 152(3): 1627.

WANG Haitao, PENG Shiwei, HE Qunyi, et al. Predicting underwater acoustic transmission loss in the SOFAR channel from ray trajectories via deep learning[J ] . JASA express letters, 2024, 4(5): 056001.

MA Fei, ZHAO Sipei, BURNETT I S. Sound field reconstruction using a compact acoustics-informed neural network[J]. The journal of the acoustical society of America, 2024, 156(3): 2009-2021.

MCCARTHY R A, MERRIFIELD S T, SARKAR J, et al. Reduced-order machine-learning model for transmission loss prediction in underwater acoustics[J]. IEEE journal of oceanic engineering, 2023, 48(4): 1149-1173.

LI Kexin, CHITRE M. Data-aided underwater acoustic ray propagation modeling[J]. IEEE journal of oceanic engineering, 2023, 48(4): 1127-1148.

YOON S, PARK Y, GERSTOFT P, et al. Predicting ocean pressure field with a physics-informed neural network[J]. The journal of the acoustical society of America, 2024, 155(3): 2037-2049.

DUAN Jie, ZHAO Hangfang, SONG Jinbao. Spatial domain decomposition-based physics-informed neural networks for practical acoustic propagation estimation under ocean dynamics[J]. The journal of the acoustical society of America, 2024, 155(5): 3306-3321.

沈远海, 马远良, 屠庆平, 等. 浅水声速剖面用经验正交函数(EOF)表示的可行性研究[J]. 应用声学, 1999, 18(2): 21-25.

SHEN Yuanhai, MA Yuanliang, TU Qingping, et al. Feasiblity of description of the sound speed profile in shallow water via empirical orthogonal functions (EOF)[J]. Applied acoustics, 1999, 18(2): 21-25.

沈远海, 马远良, 屠庆平. 声速剖面的分层经验正交函数表示[J]. 西北工业大学学报, 2000, 18(1): 90-93.

SHEN Yuanhai, MA Yuanliang, TU Qingping. On expression of ocean sound profile by layered empirical orthogonal function (EOF)[J]. Journal of Northwestern Polytechnical University, 2000, 18(1): 90-93.

何利, 李整林, 张仁和, 等. 东中国海声速剖面的经验正交函数表示与匹配场反演[J]. 自然科学进展, 2006, 16(3): 351-355.

HE Li, LI Zhenglin, ZHANG Renhe, et al. Empirical orthogonal function representation and matched field inversion of sound velocity profile in East China Sea[J]. Progress in natural science, 2006, 16(3): 351-355.

NEAL R M. Bayesian learning for neural networks[M]. New York, NY: Springer New York, 1996.

BLUNDELL C, CORNEBISE J, KAVUKCUOGLU K, et al. Weight uncertainty in neural networks[C]//In Proceedings of the 32nd International Conference on Machine Learning. Lille, 2015: 1613-1618.

黄欣, 徐荣武, 李瑞彪. 基于神经网络的船舶辐射噪声预报方法[J]. 船舶力学, 2025, 29(3): 486-496.

HUANG Xin, XU Rongwu, LI Ruibiao. Ship radiation noise prediction method based on neural network[J]. Journal of ship mechanics, 2025, 29(3): 486-496.

DEL GROSSO V A. New equation for the speed of sound in natural waters (with comparisons to other equations)[J]. The journal of the acoustical society of America, 1974, 56(4): 1084-1091.

GERSTOFT P, HUANG C F, HODGKISS W S. Estimation of transmission loss in the presence of geoacoustic inversion uncertainty[J]. IEEE journal of oceanic engineering, 2006, 31(2): 299-307.

HUANG Chenfen, GERSTOFT P, HODGKISS W S. Uncertainty analysis in matched-field geoacoustic inversions[J]. The journal of the acoustical society of America, 2006, 119(1): 197-207.

浏览量

124

下载量

CSCD

文章被引用时，请邮件提醒。

提交

工具集

关联资源

考虑系统仿真程序模型参数不确定性的RISMC方法研究

多因素影响下双时间尺度退化设备剩余寿命预测