Using NeuralPDE.jl to solve differential equations

This paper describes the application of physics-informed neural network (PINN) for solving partial derivative equations. Physics Informed Neural Network is a type of deep learning that takes into account physical laws to solve physical equations more efficiently compared to classical methods. The so...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:Discrete and continuous models and applied computational science Ročník 33; číslo 3; s. 284 - 298
Hlavní autoři: Belicheva, Daria M., Demidova, Ekaterina A., Shtepa, Kristina A., Gevorkyan, Migran N., Korolkova, Anna V., Kulyabov, Dmitry S.
Médium: Journal Article
Jazyk:angličtina
Vydáno: Peoples’ Friendship University of Russia (RUDN University) 15.10.2025
Témata:
differential equations
julia programming language
neuralpde
numerical methods
physics-informed neural networks
ISSN:2658-4670, 2658-7149
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Popis
Shrnutí:This paper describes the application of physics-informed neural network (PINN) for solving partial derivative equations. Physics Informed Neural Network is a type of deep learning that takes into account physical laws to solve physical equations more efficiently compared to classical methods. The solution of partial derivative equations (PDEs) is of most interest, since numerical methods and classical deep learning methods are inefficient and too difficult to tune in cases when the complex physics of the process needs to be taken into account. The advantage of PINN is that it minimizes a loss function during training, which takes into account the constraints of the system and th e laws of the domain. In this paper, we consider the solution of ordinary differential equations (ODEs) and PDEs using PINN, and then compare the efficiency and accuracy of this solution method compared to classical methods. The solution is implemented in the Julia programming language. We use NeuralPDE.jl, a package containing methods for solving equations in partial derivatives using physics-based neural networks. The classical method for solving PDEs is implemented through the DifferentialEquations.jl library. As a result, a comparative analysis of the considered solution methods for ODEs and PDEs has been performed, and an evaluation of their performance and accuracy has been obtained. In this paper we have demonstrated the basic capabilities of the NeuralPDE.jl package and its efficiency in comparison with numerical methods.
ISSN:2658-4670
2658-7149
DOI:10.22363/2658-4670-2025-33-3-284-298