Abstract: The study of wave-structure interactions has attracted significant attention due to its crucial role in the design, operation, and safety of marine structures. As a mesh-free numerical method, the Incompressible Smoothed Particle Hydrodynamics (ISPH) method has gradually become a promising tool for simulating wave-structure interactions. In traditional ISPH, pressure is obtained by solving the Pressure Poisson Equation (PPE), which is the most time-consuming part of the entire computational process. In this study, a hybrid method, ISPH_GNN, combining the Graph Neural Network (GNN) with ISPH, is employed to numerically simulate wave-structure interactions. In ISPH_GNN, the GNN model is used to predict fluid pressure, replacing the PPE-solving process in traditional ISPH. One of the contributions of this paper is to demonstrate that a GNN model trained on data generated from relatively simple cases can be effectively applied to more complex wave-structure interaction scenarios. Specifically, this article adopts a GNN model trained with dam-break and sloshing tank example data, and combines it with the ISPH method to construct the ISPH_GNN method, which is used to simulate various wave-structure interaction problems, including solitary wave overtopping over a step, regular wave impact on a submerged trapezoidal structure, and the interaction between regular waves and a floating box. The simulation results show that ISPH_GNN provides satisfactory results in these different scenarios, demonstrating its excellent generalization ability in wave-structure interaction problems. Another significant contribution of this study is demonstrating that, compared to the conventional ISPH method, ISPH_GNN achieves similar or slightly higher computational accuracy while significantly enhancing the efficiency of pressure prediction, especially in large-scale wave-structure interaction simulations with a high number of particles. For example, in a simulation with 1.3 million particles, the pressure prediction speed of ISPH_GNN is enhanced by up to 93 times. These results highlight the great potential of the ISPH_GNN method in wave-structure interaction simulations, offering a more scalable and computationally efficient tool for marine engineering.