An Efficient Physics-informed Residual Neural Network for Solving Seepage Squation with Source Term
This paper proposes an efficient physics-informed residual neural network(E-PIResNet)based on deep learning methods.The network constructs a unique residual structure and combines strategies such as autoregression,gradient decay,and variable time steps for model training.Numerical experiments show that this method effectively solves both homogeneous and heterogeneous problems.Compared to baseline methods,this approach reduces solution time by 50%.Additionally,this method demonstrates good robustness and generalization capabilities across examples with different permeabilities and flow rates without needing to adjust network parameters.
万方数据
维普
