The interaction between waves and offshore structures is one of the central issues in marine engineering. Traditional physical model tests are limited by cost and scale effects; therefore, the development of high-precision numerical methods is crucial for elucidating the mechanisms of fluid-structure interaction. In this study, a three-dimensional numerical wave tank was constructed based on the moving particle semi-implicit (MPS) method, combined with graphics processing unit (GPU) parallel acceleration technology, enabling high-fidelity simulation of nonlinear regular waves. Through tests of damping and wave-attenuation layer parameters, the optimal damping coefficient was determined, effectively suppressing end reflections. Numerical results indicate that the tank can stably generate preset waves with good convergence; for dimensionless wave numbers less than 2, the deviation of the average wave height from the theoretical value is less than 4%, though model performance declines beyond this range. Further simulations of the floating structure's motion response showed that the time-history curves of the caisson's motion agreed well with experimental data, verifying the model's feasibility for wave-structure interaction problems. This study provides a reliable numerical method and theoretical reference for predicting the motion of floating structures under complex wave conditions.

The Moving Particle Semi-implicit (MPS) method is a meshless method based on the Lagrangian perspective. It effectively avoids mesh distortion when simulating problems with large free-surface deformations, such as water entry and exit of structures, numerical wave generation, and dam breaks. Additionally, its free surface detection algorithm is straightforward and ensures accurate interface tracking, showing broad application prospects. To date, research on numerical wave generation using the MPS method has primarily focused on wave-structure interactions, with relatively limited validation analysis dedicated to the wave generation process itself. To address this gap, this study establishes a two-dimensional numerical wave tank model based on the MPS method, incorporating an improved particle shifting scheme and stabilization operators. Regular waves were generated using a piston-type wavemaker and numerically absorbed in a damping zone. Multiple sets of regular waves with different parameters were simulated to analyze the influence of damping coefficients on the results. Furthermore, the model was validated against experimental motion data of a floating box in waves. The results demonstrate that the proposed model can accurately simulate regular waves with various parameters. The wave amplitude error is positively correlated with the reflected wave energy, and selecting appropriate damping coefficients can effectively minimize the influence of reflected waves on the numerical results. Moreover, the simulated motion time history of the floating box agrees well with the experimental data, confirming the feasibility of the model for solving wave-structure interaction problems.