Addressing the issue of accuracy limitations of traditionalin hullship wave load inversion, this paper proposes a hybrid inversion model(PINN-LSTM) integrating Physicscal-Informed Neural Network (PINN) Information and Long Short-Term Memory Network (LSTM) is proposed.Based on finite element analysis data, the computational results of the Sesam finite element model, the collaborative framework comprising the temporatime-seriesl dynamic feature extraction, physical constraint validation, and precise hull load inversion” is established.With this framework, capability of LSTM and the physical law constraint advantage of PINN are deeply integrated to construct a collaborative inversion framework of “temporal feature extraction-physical constraint verification-precise load inversion”.The LSTM is employed to module extracts the temporal feature froms of multi-point strain sequences, while PINN ensures physical consistency by embedding the constitutive relationship and boundary condition constraints based on the hull bending and torsion theory, achieving joint optimization of data fitting accuracy and physical rationality.Experimental results show that the proposed model significantly outperforms the traditional Moore-Penrose inverse model, standalone LSTM and PINN models in predicting vertical bending moment, lateral bending moment, and torque across various sections and wave headings.Notably, the model under different conditions such as multiple profiles and wave directions, the inversion accuracy of this model in terms of vertical bending moment, lateral bending moment, and torque, three wave load components, is significantly better than that of traditional methods using Moore-Penrose inversion, pure LSTM models, and pure PINN models.It maintains high accuracy and physical consistency in complex scenarios such as sudden load mutations changes and bending-torsion coupling.Ablation studies experiments and generalization ability analysis further verify the effectiveness and cross-condition adaptability of the models in the proposed framework,each module of the model.This study not only provides a novel high-precison approach for hull wave load inversion and a new method for efficient and high-precision inversion of ship wave loads but also offers a transferable technical paradigm for solving temporal inverse problems in complex physical systems.
波浪载荷反演任务旨在基于船体结构上布设的应变传感器所监测到的多点应变数据,反演出作用于船体横剖面的三类主要动态载荷分量:垂向弯矩(反映船体在波浪中上下弯曲的强度)、横向弯矩(反映左右弯曲的强度)以及自由扭转扭矩(反映船体绕纵轴扭转变形的强度)[1]。当前关于船舶动态响应识别的研究主要集中于时域和频域计算方法。时域方法是通过直接求解微分方程,研究系统状态或载荷随时间变化的完整过程[2]。Ren等[3]通过对船体结构施加特定载荷获得响应,建立影响系数矩阵,获得动态载荷。Jiang等[4]提出了一种基于 Newmark-β方法的时域算法,以解决连续系统动态荷载识别问题。但目前时域方法识别结果的准确性高度依赖于采样时间间隔,实际精度仅在较小间隔内有效,且不可避免地会产生规模巨大的核函数响应矩阵。在频域计算方法上,Kang等[5]将动态荷载转化为等效静态荷载集,并使结构在某一时刻的位移场与动态荷载产生的位移场相同。李军等[6]基于频域随机动载荷识别理论,采用 Moore Penrose广义逆方法,识别海上导管架风机的气动载荷。目前基于时域频域的动态响应识别方法对计算要求较高,识别精度依赖于响应信息的准确性[7],且本质上是一个病态的逆问题,微小测量误差可能导致解的巨大偏差。
ZhangM, SunL, XieY.A monitoring method of hull structural bending and torsional moment[J].Ocean Eng, 2024, 291: 116344.
[2]
LeeY W, CheathamT P, WiesnerJ B.Application of correlation analysis to the detection of periodic signals in noise[J].Proc IRE, 1950, 38(10): 1165-1171.
[3]
RenH L, FengG Q, LiuH, et al.Dynamic load inversion method of ship body based on influence coefficient matrix[C]//ASME 2019 38th International Conference on Ocean, Offshore and Arctic Engineering, 2019.
[4]
JiangJ, DingM, LiJ.A novel time-domain dynamic load identification numerical algorithm for continuous systems[J].Mech Syst Signal Process, 2021, 160: 107881.
[5]
KangB S, ChoiW S, ParkG J.Structural optimization under equivalent static loads transformed from dynamic loads based on displacement[J].Comput Struct, 2001, 79(2): 145-154.
[6]
LiJ, YuZ F, ChenS G, et al.Study on structural model modification and load identification based on limited measuring points[J].Chin J Ship Res, 2023(5): 57-64.
HuY Y.Application research of force identification based on least squares method in the frequency domain[D].Harbin: Harbin Engineering University, 2011.
[9]
胡寅寅.基于频域最小二乘的载荷识别方法与应用研究[D].哈尔滨: 哈尔滨工程大学, 2011.
[10]
WangJ, HanL, DongC, et al.A PINN-based wave model for prediction of wave period and wavelength using dispersion relation[J].Ocean Eng, 2025, 341: 122504.
[11]
LiuR, LiH, OngM C, et al.Prediction of global structural responses for an ultra-large container ship under long-crested irregular waves using generalized modes and artificial neural networks[J].Ocean Eng, 2025, 318: 120152.
[12]
SunH, JiaoJ, XuW, et al.Intelligent prediction of ship wave loads and whipping responses from input motions by decomposition-superposition method[J].Ocean Eng, 2026, 343: 123267.
[13]
Álvarez ChavesM, Acuña EspinozaE, EhretU, et al.When physics gets in the way: An entropy-based evaluation of conceptual constraints in hybrid hydrological models[J].Hydrol Earth Syst Sci, 2026, 30(3): 629-658.
[14]
JiangZ T, HuangM Z, PengY X, et al.Engineering application and experimental validation of the FEM-PIKAN model in ship hull deformation analysis [J].Ocean Eng, 2026, 344: 123588.
[15]
SuX, HuangY, ZhouL, et al.Innovative load identification with Res-UNet: Integrating phase space reconstruction and physics-informed deep learning[J].Ocean Eng, 2024, 312: 119173.
[16]
SunL, ZhuB, QinZ, et al.Intelligent ship seakeeping prediction using physics-informed neural networks [C]//2025 6th International Conference on Mechatronics Technology and Intelligent Manufacturing (ICMTIM), 2025: 578-583.
[17]
ZhangM, QiuB, ZhuM, et al.Novel computation method of reducing ill-posedness for structural static distributed load identification by optimising strain gauge locations[J].Mech Syst Signal Process, 2019, 124: 83-110.
[18]
LiuY J.Study on ship wave load inversion method based on stress monitoring data[D].Harbin: Harbin Engineering University, 2023.
[19]
刘玉洁.基于应力监测数据的船舶波浪载荷反演方法研究[D].哈尔滨: 哈尔滨工程大学, 2023.
[20]
ZhangM.The indirect monitoring method of typical loads exerted on ship and marine structures[D].Harbin: Harbin Engineering University, 2020.
[21]
张猛.船舶与海洋工程结构典型载荷间接监测方法研究[D].哈尔滨: 哈尔滨工程大学, 2020.
[22]
NielsenU D, BinghamH B, BrodtkorbA H, et al.Estimating waves via measured ship responses[J].Sci Rep, 2023, 13: 17342.