Author(s): Yangsha Ye; Zhengzhi Deng; Xin Bian; Zijun Hu
Linked Author(s): Zhengzhi Deng
Keywords: Physics-information neural networks terrain inversion wave field reconstruction
Abstract: Accurate acquisition of Marine topography data is essential for understanding the movement of seabed plates and seabed evolution, which is of great significance to resource development, marine disaster prevention, military defense, etc. Traditional submarine topography detection technology has problems such as low measurement accuracy, insufficient resolution, and narrow coverage. To solve this problem, this paper proposes a nearshore terrain inversion method based on physics-information neural networks (PINNs). By embedding wave energy balance equation and dispersion relationship into the neural network, the classical finite element method is used to obtain hydrodynamic data as the marker data for model training and the reference for prediction, so as to verify the accuracy of the model. The inversion of different seabed landforms with sparse data is realized. It is proved that the model can reconstruct the wave field effectively using a small amount of wave height measurement data, and achieve high-precision terrain inversion with relative error and maximum error less than 5%.
Year: 2025