Physical-Guided Coupling Neural Network Approach for Seismic Wave Propagation

Seismic wave propagation is mainly studied by two paradigms: empirical research based on in-situ observation and model test, theoretical research based on mathematical deduction and numerical simulation. However, these paradigms face challenges such as sparse data samples, weak generalization of results, and insufficient understanding of laws. To address these challenges, we propose a coupling neural network that embeds both physical information and constrains physical laws. We use this neural network to learn the law of seismic wave propagation from a combination of theoretical equations and test records. We develop a prediction model of seismic wave propagation that jointly constrains multi-type sparse data, which improves the physical interpretability and extrapolation ability. The results demonstrate that the physical-guided coupling neural network can effectively and flexibly integrate theoretical, simulated, and experimental data, and generate the full waveform data and spatial distribution patterns of various physical quantities, thereby reducing the uncertainty of sparse sensor test data and solving the problem of data interaction of independent research paradigms.