Shallow water equations-fused dam-break wave propagation prediction model ensembled with a training process resampling method

The prediction of dam-break wave propagation behavior is a long-standing problem in hydrodynamics and hydrology. In addition to the traditional numerical simulation methods, deep learning technology has gradually made some beneficial attempts in the exploration of dam-break wave propagation. However, pure data-driven deep learning methods have certain limitations in terms of dataset requirements and long-term prediction. Therefore, with the help of the idea of the physical information neural networks (PINNs), this paper proposes a dam-break wave prediction neural network model embedded with the shallow water equations (SWEs) ——SWENet model. At the same time, in order to overcome the problem of low computational efficiency caused by large-scale computational domain and better capture the characteristics of the dam-break wave, a training progress random resampling (TPRR) method is proposed. The effectiveness of the SWENet model is verified by using two-dimensional dam-break cases, and compared with the pure data-driven multilayer perceptron (MLP) model. The results show that the SWENet model can achieve long-term dam-break wave prediction with less training data. We also analyze the training data size N
s
, N
e
and the resampling parameter E
e
, and give the appropriate values.