Predicting waves in fluids with deep neural network

In this paper, we present a deep learning technique for data-driven predictions of wave propagation in fluid medium. The relies on an attention-based convolutional recurrent autoencoder network (AB-CRAN). To construct low-dimensional representation data, employ denoising-based autoencoder. AB-CRAN architecture with long short-term memory cells forms our neural model the time marching features. We assess proposed framework against standard propagation. demonstrate effectiveness model, consider three benchmark problems, namely, one-dimensional linear convection, nonlinear viscous Burgers equation, and two-dimensional Saint-Venant shallow water system. Using spatial-temporal datasets from novel accurately captures amplitude preserves characteristics solution horizons. sequence-to-sequence increases time-horizon prediction compared to cells. denoising further reduces mean squared error improves generalization capability parameter space.