Predicting ocean pressure field with a physics-informed neural network.

Area of Science:

• Ocean acoustics
• Machine learning
• Computational physics

Background:

• Ocean sound pressure fields are complex and challenging to predict, especially over long ranges.
• Traditional methods struggle with rapidly fluctuating phases in acoustic data.
• Accurate sound field prediction is crucial for underwater acoustics and sonar applications.

Purpose of the Study:

• To develop a novel machine learning strategy for predicting ocean sound pressure fields.
• To enhance the accuracy and efficiency of acoustic field prediction in complex ocean environments.
• To leverage physics-informed neural networks (PINNs) for improved data-driven modeling.

Main Methods:

• Utilizing a physics-informed neural network (PINN) that integrates acoustic data with governing partial differential equations (PDEs).
• Employing the envelope function from the parabolic-equation technique to handle phase variations.
• Training the neural network with range-depth data to predict complex acoustic pressure.

Main Results:

• The proposed PINN strategy effectively predicts sound pressure fields in ocean waveguides.
• The use of the envelope function significantly improves neural network convergence and accuracy.
• PINNs demonstrate capability in capturing PDE solutions even with limited training data.

Conclusions:

• Physics-informed neural networks offer a powerful approach for ocean sound pressure field prediction.
• Integrating physical laws with machine learning enhances predictive capabilities in complex acoustic environments.
• The method shows promise for real-world applications, validated by simulations and experimental data.