Research on Intelligent Prediction Method of Underwater Acoustic Field Based on Parabolic Equation

In the field of hydroacoustic field prediction, numerical simulation methods and machine learning techniques are two commonly used methods. However, numerical simulation methods need to be meshed and are computationally intensive in the case of complex models; machine learning methods are difficult to analyze the physical significance of the model. Aiming at the above problems, combined with the parabolic equation theory, this paper explores the feasibility of two new neural network models for physical processes for acoustic field prediction. Using the physical information neural network (PINN) and Fourier neural operator (FNO), the coordinates of each point of the hydroacoustic field and the corresponding sound pressure are taken as inputs, and the predicted value of the sound pressure at the requested location is output. The RAMv1.5 model is utilized to generate sample data, and the training and testing are carried out using data from different sound sources, and the simulation results conclude that both models are effective in predicting the distribution of the underwater sound field.