Wave simulation of symmetric V-shaped canyon based on physics-informed deep learning method

The seismic effects of alpine and canyon sites are a research hotspot in the field of earthquake engineering. The series solution of the two-dimensional scattering and diffraction wave functions of cylindrical SH waves caused by V-shaped canyons is relatively mature and provides reasonable and scientific ground motion input for many major projects in river valleys. In this study, the physics-informed deep learning method combined with the comparative analysis of the analytical results is used to further clarify the seismic response characteristics and complex wave field spatial distribution of the V-shaped river valley. The method mainly focuses on sparse samples and interpretable artificial intelligence, and establishes a deep neural network to realize the semi-infinite seismic propagation model by combining the strong formal automatic differentiation with the soft constraint boundary condition embedding, and realizes high-precision prediction of V-shaped river valleys under different given wave field conditions by adopting the time domain decomposition strategy. By comparing with the analytical solution, the accuracy and efficiency of the proposed physics-driven artificial intelligence method are evaluated. The results show that the physics-driven artificial intelligence method can be applied to the analysis of terrain effects, and the cylindrical SH waves are significantly attenuated at the bottom of the V-shaped canyon, and the edge area shows an amplification effect.