Abstract:This paper presents the double-activation neural network (DANN), a novel network architecture designed for solving parabolic equations with time delay. In DANN, each neuron is equipped with two activation functions to augment the network's nonlinear expressive capacity. Additionally, a new parameter is introduced for the construction of the quadratic terms in one of two activation functions, which further enhances the network's ability to capture complex nonlinear relationships. To address the issue of low fitting accuracy caused by the discontinuity of solution's derivative, a piecewise fitting approach is proposed by dividing the global solving domain into several subdomains. The convergence of the loss function is proven. Numerical results are presented to demonstrate the superior accuracy and faster convergence of DANN compared to the traditional physics-informed neural network (PINN).

What problem does this paper attempt to address?

This paper proposes a new method called Double-Activation Neural Network (DANN) for solving parabolic partial differential equations (PDDEs) with time delay. Traditional methods often have high computational cost and low accuracy when dealing with problems with state-dependent delays due to the discontinuity of the solutions. DANN enhances the nonlinear expression ability of the network by equipping each neuron with two activation functions and introduces new parameters to construct quadratic terms, thus capturing complex nonlinear relationships better. The paper first introduces the applications of delay differential equations and delay partial differential equations in biology, control, climate prediction, and other fields. Then, the authors propose an improvement strategy for Physics-Informed Neural Networks (PINN), which includes DANN and piecewise fitting methods. DANN improves the approximation accuracy of the network to the solution by adding two activation functions and a new parameter for constructing quadratic terms in each neuron. The piecewise fitting method divides the entire domain into multiple subdomains to deal with discontinuity points in the solution caused by nonzero delays and improves the fitting accuracy. In addition, the paper demonstrates the convergence of the loss function and shows the advantages of DANN over traditional PINN, Adaptive PINN (APINN), Quadratic Residual Networks (QRES), Identity Shortcut Networks (ISC), and Quadratic Shortcut Networks (QSC) in terms of accuracy and convergence speed through numerical experiments. The experimental results show that DANN achieves the highest fitting accuracy with fewer training points. In conclusion, the main contributions of this paper are the proposal of a new neural network architecture, DANN, and a piecewise fitting strategy for dealing with delay partial differential equations. These methods effectively solve complex problems with state-dependent delays and improve the efficiency and accuracy of numerical solutions.

Double-activation neural network for solving parabolic equations with time delay

A Physics Informed Neural Network for Time-Dependent Nonlinear and Higher Order Partial Differential Equations

Densely Multiplied Physics Informed Neural Networks

Mass-preserving Spatio-temporal adaptive PINN for Cahn-Hilliard equations with strong nonlinearity and singularity

Adaptive quadratures for nonlinear approximation of low-dimensional PDEs using smooth neural networks

Data-driven solutions and parameter estimations of a family of higher-order KdV equations based on physics informed neural networks

Physical informed neural networks with soft and hard boundary constraints for solving advection-diffusion equations using Fourier expansions

A Dimension-Augmented Physics-Informed Neural Network (DaPINN) with High Level Accuracy and Efficiency

NAS-PINN: Neural architecture search-guided physics-informed neural network for solving PDEs

Advanced Physics-Informed Neural Networks for Numerical Approximation of the Coupled Schrödinger-Kdv Equation

Multi-Net Strategy: Accelerating Physics-Informed Neural Networks for Solving Partial Differential Equations.

Physics-informed neural networks with parameter asymptotic strategy for learning singularly perturbed convection-dominated problem

TSONN: Time-stepping-oriented neural network for solving partial differential equations

Efficient Discrete Physics-informed Neural Networks for Addressing Evolutionary Partial Differential Equations

Physics-informed neural networks based on adaptive weighted loss functions for Hamilton-Jacobi equations

Development of backward compatible physics-informed neural networks to reduce error accumulation based on a nested framework

Physics-Informed Neural Networks for Solving High-Index Differential-Algebraic Equation Systems Based on Radau Methods

PPINN: Parareal physics-informed neural network for time-dependent PDEs

Solving Partial Differential Equations Using Deep Learning and Physical Constraints