融合物理感知加权机制的图神经网络及其在海床冲蚀预测中的应用

刘少腾; 赵婷婷; 田浩; 李文杰; 李志强; 冯云田

doi:10.12382/bgxb.2025.0999

您当前的位置：

首页 >

文章列表页 >

融合物理感知加权机制的图神经网络及其在海床冲蚀预测中的应用

更新时间：2026-02-16

- 融合物理感知加权机制的图神经网络及其在海床冲蚀预测中的应用
- A Graph Neural Network with Physics-Aware Weighting and Its Application in Seabed Scour Prediction
- 兵工学报 2026年
- 作者机构：
  
  1. 太原理工大学航空航天学院,山西,太原,030024
  2. 材料强度与结构冲击山西省重点实验室,山西,太原,030024
  3. 斯旺西大学辛克维奇计算工程中心,斯旺西,英国,SA1 8EN
- 作者简介：
- 基金信息：
  
  国家自然科学基金项目(12102294)
- DOI：10.12382/bgxb.2025.0999
  中图分类号： O358
- 收稿：2025-11-05，
  
  网络首发：2026-02-13，
- 稿件说明：
移动端阅览
刘少腾，赵婷婷，田浩，等. 融合物理感知加权机制的图神经网络及其在海床冲蚀预测中的应用[J/OL]. 兵工学报, 2026(2026-02-16). https://doi.org/10.12382/bgxb.2025.0999.

LIU S T, ZHAO T T, TIAN H, et al. A graph neural network with physics-aware weighting and its application in seabed scour prediction[J/OL]. Acta Armamentarii, 2026(2026-02-16). https://doi.org/10.12382/bgxb.2025.0999. (in Chinese)
刘少腾，赵婷婷，田浩，等. 融合物理感知加权机制的图神经网络及其在海床冲蚀预测中的应用[J/OL]. 兵工学报, 2026(2026-02-16). https://doi.org/10.12382/bgxb.2025.0999. DOI：

LIU S T, ZHAO T T, TIAN H, et al. A graph neural network with physics-aware weighting and its application in seabed scour prediction[J/OL]. Acta Armamentarii, 2026(2026-02-16). https://doi.org/10.12382/bgxb.2025.0999. (in Chinese) DOI：

摘要

准确预测水射流冲击海床形成的冲蚀坑深度，对海底光缆埋设工艺的参数优化与设备选型具有重要意义。传统数值模拟方法虽精度较高，但计算成本高昂、耗时较长。图神经网络（Graph Neural Networks

GNN）在模拟大变形物理过程中表现出色，但在处理局部小变形问题时精度有限。为此，本文提出一种融合物理感知的大变形加权机制（Pit-Aware Weighting

PAW）的图神经网络模型（PAW-GNN），通过将海床结构离散为粒子系统，构建图结构并引入自回归预测机制，实现对水射流冲击下海床坑深动态演化的高效预测。模型在包含40组仿真案例的数据集上进行了训练与验证，涵盖不同射流宽度与速度工况。测试结果表明，PAW-GNN模型在预测精度上与传统数值模拟结果一致，平均绝对百分比误差与决定系数指标达到了较好的水平；在计算效率上较传统数值方法提升一个数量级；在插值与外推测试中展现出良好的泛化能力。该模型为水射流冲击海床过程的快速预测提供了一种可靠的替代方案。

Abstract

Accurate prediction of the scouring crater depth formed by water jet impingement on the seabed is of great significance forparameter optimization and equipment selection in submarine cable burial engineering. Although traditional numerical simulation methods offer high accuracy

they suffer from high computational costs and long processing times. Graph Neural Networks (GNNs) excel at simulating large-deformation physical processes but exhibit limited accuracy in handling localized small-deformation problems. To address this

this paper proposes a GNN model integrated with a physics-aware large-deformation weighting mechanism

termed PAW-GNN. By discretizing the seabed structure into a particle system

constructing a graph representation

and employing an autoregressive prediction strategy

the model efficiently predicts the dynamic evolution of seabed scour depth under water jetimpact. The model was trained and validated on a dataset comprising 40 simulation cases covering various jet widths and velocities. The test results demonstrate that the PAW-GNN model achieves high consistency with traditional numerical simulations in terms of prediction accuracy

with the mean absolute percentage error and coefficient of determination reaching satisfactory levels. Furthermore

it improves computational efficiency by an order of magnitude compared to conventional numerical methods. The model also exhibits robust generalization capability in both interpolation andextrapolation tests. This model provides a reliable alternative for the rapid prediction of water jet-induced seabed scouring processes.

关键词

Keywords

references

谢永顺,张琦琦,王成金. 全球海底光缆网络的脆弱性结构及区域抗毁性评价[J]. 地理科学,2024,44(3): 411-420.XIE Y S, ZHANG Q Q, WANG C J. Vulnerability structure of the global submarine optical cable network and regional invulnerability assessment[J]. Scientia Geographica Sinica, 2024, 44(3): 411-420.(in Chinese)

武琼, 蒲婧新. 中美在海底光缆领域的战略竞争及影响[J]. 和平与发展, 2022(4): 81-102, 137-138. WU Q, PU J X. Strategic competition between china and the united states in the submarine optical cable domain and its implications[J]. Peace and Development, 2022(4): 81-102, 137-138.(in Chinese)

张安安,何明光,于璐,等. 基于有限元法的海底电缆埋设深度实时预测研究[J]. 电子测量与仪器学报,2023,37(12): 18-28. ZHANG A A, HE M G, YU L, et al. Real-time prediction of submarine cable burial depth based on the finite element method[J]. Journal of Electronic Measurement and Instrumentation, 2023, 37(12): 18-28.(in Chinese)

林如. 海底管道射流式挖沟机的研究及射流破土的仿真分析[D]. 哈尔滨工程大学,2019. LIN R. Research on submarine pipeline jet trenching machine and simulation analysis of soil breaking by water jets[D]. Harbin Engineering University, 2019.(in Chinese)

曹淑刚,张吉,孙小钎,等. 考虑弯曲刚度的高压海缆敷设受力分析[J]. 太阳能学报,2019,40(10): 3009-3016. CAO S G, ZHANG J, SUN X Q, et al. Force analysis of high-voltage submarine cable laying considering bending stiffness[J]. Acta Energiae Solaris Sinica, 2019, 40(10): 3009-3016.(in Chinese)

QIU Y Q, LAN X D, YANG Z P, et al. Study on themechanismof soil erosion by submerged water jet vertical scouring in cohesive soils[J]. Ocean Engineering, 2024, 311: 118919.

ALHADDAD S, SNYDER A, BULT S V, et al. Experimental investigation of cohesive soil erosion caused bytranslating submerged inclined water jets[J]. Applied Ocean Research, 2025, 157: 104486.

LI H, LI J B, HUANG Z W, et al. Numerical investigation of abrasive water jet impact force characteristics by using the coupled SPH-FEM method: considering the shape, the interaction and the fragmentation of abrasive particles[J]. Powder Technology, 2025: 121287.

ZHAO H H, JIANG H X, WARISAWA S, et al. Numerical study of abrasive water jet rotational slits in hard rock usinga coupled SPH-FEM method[J]. Powder Technology, 2023, 426: 118622.

REN F S, FANG T C, CHENG X Z. Study on rock damage and failure depth under particle water-jet coupling impact[J].International Journal of Impact Engineering, 2020, 139:103504.

WANG W, WANG C, YANG T, et al. A friction interface model for multi-material interactions in a Eulerian framework[J]. Journal of Computational Physics, 2021, 433:110057.

LI Z T, GE Z L, ZHOU Z, et al. Numerical simulation and experimental verification of heterogeneous granite impacted by abrasive water jet based on SPH-FEM coupling algorithm[J]. Powder Technology, 2023, 416: 118233.

GONG W J, WANG J M, GAO N. Numerical simulation for abrasive water jet machining based on ALE algorithm[J]. The International Journal of Advanced Manufacturing Technology, 2011, 53(1): 247-253.

王金涛, 余文力, 王涛, 等. 异形炸药部件的水射流安全分解技术[J]. 兵工学报, 2020, 41(11): 2216-2224. WANG J T, YU W L, WANG T, et al. Safe disassembly of complex-shaped high-explosive components using water jet technology[J]. Acta Armamentarii, 2020, 41(11): 2216-2224.(in Chinese)

ZHOU W J, WANG X J, LIU B, et al. Machine learning method to predict solid propellant breakage efficiency of cavitation water jet[J]. Propellants, Explosives, Pyrotechnics, 2022, 47(12): e202200131.

COJBASIC Z, PETKOVIC D, SHAMSHIRBAND S, et al. Surfaceroughness prediction by extreme learning machine constructed with abrasive water jet[J]. Precision Engineering,2016, 43: 86-92.

LI Q L, WANG Z T, LI L, et al. Machine learning prediction of structural dynamic responses using graph neural networks[J]. Computers & Structures, 2023, 289: 107188.

SCARSELLI F, GORI M, TSOI A C, et al. The graphneuralnetwork model[J]. IEEE transactions on neural networks, 2008, 20(1): 61-80.

LUO M, DOU H L, ZHENG N. Spatiotemporal prediction of urban traffics based on deep GNN[J]. Computers, Materials & Continua, 2024, 78(1).

HOSHEN Y. Vain: Attentional multi-agent predictive modeling[J]. Advances in neural information processing systems, 2017, 30.

LI Q L, WANG Y, CHEN W S, et al. Machine learning prediction of BLEVE loading with graph neural networks[J].Reliability Engineering & System Safety, 2024,241: 109639.

LI Q L, WANG Z T, CHEN W S, et al. Advancing blast fragmentation simulation of RC slabs: A graph neural network approach[J]. Engineering Structures, 2024, 308: 118009.

GAO R, JAIMAN R K. Quasi-monolithic graph neural network for fluid-structure interaction[J]. CoRR, 2022.

彭江舟, 潘刘娟, 高光发, 等. 城市建筑外爆威力场与毁伤效应数智仿真模型及应用[J]. 爆炸与冲击, 2025. PENG J Z, PAN L J, GAO G F, et al. Digital-intelligent simulation model and application of blastfield and damage effects of urban building external explosions[J]. Explosion and Shock Waves, 2025.(in Chinese)

LIN C, XIE Y, WANG C H. Significant wave height prediction at multiple sites using sequence decomposition and dynamic spatiotemporal graph neural networks[J].Ocean Engineering, 2025, 341: 122548.

KIM J, KIM T, CHANG S, et al. Prediction of beach profile changes using spatiotemporal graph neural networks for beach morphological evolution modeling[J]. Ocean Engineering, 2025, 340: 122282.

池寅, 时豫川, 吴海洋, 等. 水射流冲埋砂质海床土体数值模拟[J]. 振动与冲击, 2019, 38(13): 253-260. CHI Y, SHI Y C, WU H Y, et al. Numerical simulation of sandy seabed soils eroded by water jets[J]. Journal of Vibration and Shock, 2019, 38(13): 253-260.(in Chinese)

温赞江. 水刀水射流速度计算方法[J]. 机械工程师, 2011(3): 50-51.

WEN Z J. Calculation method of water jet velocity for water jet[J]. Mechanical Engineer, 2011(3): 50-51. (in Chinese)

Feng C, Kong L, Wang Y, et al. Experimental study on scouring by submerged pulsed waterjet vertically impinging on cohesive bed[J]. Ocean Engineering, 2024, 298: 117308.

ORTOLANI C, IRVINE J, BURLEY L, et al. Framework for jet trenching performance prediction in cohe-sionless soils[J]. 2025.

Terzaghi K, Peck R B, Mesri G. Soil mechanics in engineering practice[M]. New York: John Wiley & Sons, 1996: 174.

浏览量

下载量

CNKI被引量

文章被引用时，请邮件提醒。

提交

工具集

关联资源

暂无数据