In recent years, deep learning approaches for partial differential equations have received much attention due to their mesh-freeness and other desirable properties. However, most of the works so far concentrated on time-dependent nonlinear differential equations. In this work, we analyze potential issues with the well-known Physic Informed Neural Network for differential equations that are not time-dependent. This analysis motivates us to introduce a novel technique, namely FinNet, for solving differential equations by incorporating finite difference into deep learning. Even though we use a mesh during the training phase, the prediction phase is mesh-free. We illustrate the effectiveness of our method through experiments on solving various equations.
翻译:近年来,部分差异方程式的深层次学习方法因其网状自由性和其他可取特性而受到很大关注。然而,迄今为止,大部分工程都集中在时间上的非线性差异方程式上。在这项工作中,我们分析了与著名的物理、知情神经网络对不依赖时间的不同方程式的潜在问题。这一分析促使我们引入一种新型技术,即FinNet,通过将有限差异纳入深层次学习来解决差异方程式。尽管我们在培训阶段使用了网状,但预测阶段是无网状的。我们通过解决各种方程式的实验展示了我们方法的有效性。