2025_programme: Physics-informed neural networks (PINNs) for underwater acoustic propagation modeling: A review
- Day: June 17, Tuesday
Location / Time: D. CHLOE at 15:30-15:50
- Last minutes changes: -
- Session: 06. Enhancing underwater acoustic sensing through machine learning
Organiser(s): Cheng Chi
Chairperson(s): Peng Xiao
- Lecture: Physics-informed neural networks (PINNs) for underwater acoustic propagation modeling: A review
Paper ID: 2366
Author(s): Peng Xiao, Yuxiang Gao, Zhenglin Li
Presenter: Yuxiang Gao
Abstract: In the field of solving partial differential equations (PDE), physical information neural networks (PINNs) have become a subject of considerable interest. The core of underwater acoustic modeling is to obtain the solution to the wave equation; therefore, PINNs have considerable potential for application in this domain. This talk provides a comprehensive review of the latest research progress of PINNs in underwater acoustic propagation modeling, which can be broadly categorized into two types. The first of these involves the introduction of certain reasonable assumptions through mathematical methods to simplify the wave equation, with PINNs then being utilized to find solutions. Examples of this include ray-assisted PINNs and the use of PINNs to estimate modal wave numbers. The second category of research focuses on optimizing the hyperparameters and architecture of neural networks to enhance performance, with examples of this being the design of decoupled neural networks and spatial domain decomposition methods. The extant research foundation indicates that PINNs demonstrate substantial potential in the computation and modeling of underwater acoustic propagation; however, there are still certain limitations in terms of computational speed and accuracy. In the future, a highly promising application direction of PINNs is to invert ocean acoustic environmental parameters using a small amount of measured data obtained from experiments and conduct acoustic field reconstruction and expansion.
- Corresponding author: Prof Peng Xiao
Affiliation: School of Ocean Engineering and Technology, Sun Yat-sen University
Country: China
