Physics-Informed Neural Networks for Solving Underwater Two-dimensional Sound Field

This paper proposes a fast computational method for underwater acoustic fields based on physics-informed neural networks (PINNs). By encoding the Helmholtz equation in the frequency domain, which represents the propagation laws of underwater sound, into physical neurons using automatic differentiation, and combining a small amount of measured data as information neurons, a physics-informed and data-driven neural network was constructed to solve the underwater acoustic fields. Simulation results demonstrate that the proposed PINNs model can effectively solve the low-frequency shallow-water acoustic field computation problem. The results of our model are highly consistent with those obtained from traditional numerical models. Compared to purely data-driven models, the PINN model shows significant generalization ability and robustness, indicating promising engineering applications.