TCAS-PINN: Physics-informed neural networks with a novel temporal causality-based adaptive sampling method
Jia Guo,Haifeng Wang,Shilin Gu,Chenping Hou
DOI: https://doi.org/10.1088/1674-1056/ad21f3
2024-01-25
Chinese Physics B
Abstract:Physics-informed neural networks (PINNs) have become a kind of attractive machine learning framework for obtaining solutions of partial differential equations (PDEs). PINNs embed initial, boundary, and PDE constraints into the loss function. The performance of PINNs is generally affected by both training and sampling. Specifically, training methods focus on how to overcome the training difficulties caused by the special PDE residual loss of PINNs, and sampling methods are related to the location and distribution of sampling points on which evaluation of PDE residual loss is accomplished. However, a common problem among these original PINNs is that they omit special temporal information utilization in neither training nor sampling levels when facing the important PDE category, i.e., time-dependent PDEs, where temporal information is the key point for algorithms. There is one work called causal PINN that considers temporal causality at the training level but lacks special temporal utilization at the sampling level. Incorporating temporal knowledge into sampling is still to be studied. To fill this gap, we propose a novel temporal causality-based adaptive sampling method that dynamically determines the sampling ratio according to both PDE residual and temporal causality. By designing a sampling ratio determined by both residual loss and temporal causality to control the number and location of sampled points in each temporal subdomain, we provide a practical solution by incorporating temporal information into sampling. Numerical experiments of several nonlinear time-dependent PDEs, including Cahn-Hilliard, KdV, Allen-Cahn, and Wave equations, show that our proposed sampling method can improve performance. We demonstrate that using such a relatively simple sampling method can improve prediction performance up to two orders of magnitude compared with other results, especially when points are limited.
physics, multidisciplinary