Abstract:In this paper, we employ the Physics-Informed Neural Network (PINN) to estimate the practical range of the characteristic wavelength parameter(referred to as the smoothing parameter) $\alpha$ in the Leray-Burgers equation. The Leray-Burgers equation, a regularization of the inviscid Burgers equation, incorporates a Helmholtz filter with a characteristic wavelength $\alpha$ to replace the usual convective velocity, inducing a regularized convective velocity. The filter bends the equation's characteristics slightly and makes them not intersect each other, leading to a global solution in time. By conducting computational experiments with various initial conditions, we determine the practical range of $\alpha >0$ that closely approximates the solutions of the inviscid Burgers equation. Our findings indicate that the value of $\alpha$ depends on the initial data, with the practical range of $\alpha$ being between 0.01 and 0.05 for continuous initial profiles and between 0.01 and 0.03 for discontinuous initial profiles. The Leray-Burgers equation captures shock and rarefaction waves within the temporal domain for which training data exists. However, as the temporal domain extends beyond the training interval, data-driven forward computation demonstrates that the predictions generated by the PINN start to deviate from the exact solutions. This study also highlights the effectiveness and efficiency of the Leray-Burgers equation in real practical problems, specifically Traffic State Estimation.

What problem does this paper attempt to address?

### Problems the Paper Attempts to Solve This paper attempts to use Physics-Informed Neural Networks (PINN) to estimate the practical range of the characteristic wavelength parameter (also known as the smoothing parameter) α in the 1D Leray-Burgers equation. Specifically, the Leray-Burgers equation is a regularization of the inviscid Burgers equation by introducing a Helmholtz filter with a characteristic wavelength α to replace the usual convective velocity, thereby producing a regularized convective velocity. This filter slightly bends the characteristics of the equation, preventing them from intersecting, thus obtaining a global solution over time. The main objectives of the paper include: 1. **Determining the practical range of α**: By conducting computational experiments under various initial conditions, determine the practical range of α that can approximate the solution of the inviscid Burgers equation. 2. **Validating the effectiveness of the Leray-Burgers equation**: Study the effectiveness and efficiency of the Leray-Burgers equation in practical problems, particularly in Traffic State Estimation (TSE). 3. **Analyzing the predictive capability of PINN**: Explore the predictive capability of PINN in different time domains, especially its performance beyond the training time domain. ### Main Findings - **Continuous initial profiles**: For continuous initial profiles, the practical range of α is between 0.01 and 0.05. - **Discontinuous initial profiles**: For discontinuous initial profiles, the practical range of α is between 0.01 and 0.03. - **Long-term prediction**: As the time domain extends beyond the training interval, the predictions generated by PINN start to deviate from the exact solution, indicating the need for more data or improved model architecture for long-term predictions. - **Traffic State Estimation**: The Leray-Burgers equation performs excellently in traffic state estimation, outperforming the diffusion-based LWR-ϵ model with higher computational efficiency. ### Experimental Methods - **PINN architecture**: A Multilayer Perceptron (MLP) is used as the base neural network, combined with physical constraints to train the model. - **Dataset**: Datasets are generated through numerical or analytical methods for different initial and boundary conditions of the inviscid and viscous Burgers equations. - **Optimization methods**: The L-BFGS optimizer is used for training, with the ADAM optimizer used for preprocessing when necessary. ### Conclusion This study successfully utilized PINN to estimate the practical range of the characteristic wavelength parameter α in the Leray-Burgers equation and validated the effectiveness and efficiency of this equation in practical problems. However, the performance of PINN in long-term predictions still needs improvement, especially in addressing oscillation issues when dealing with discontinuities. These results provide important references for the application of the Leray-Burgers equation in practical scenarios such as traffic state estimation.

Estimation of the Characteristic Wavelength Parameter in 1D Leray-Burgers Equation with PINN

A Combination of Deep Neural Networks and Physics to Solve the Inverse Problem of Burger's Equation

Physics-Informed Neural Networks with Adaptive Localized Artificial Viscosity

Singular layer PINN methods for Burgers' equation

Solving Viscous Burgers' Equation: Hybrid Approach Combining Boundary Layer Theory and Physics-Informed Neural Networks

Investigating the Ability of PINNs To Solve Burgers' PDE Near Finite-Time BlowUp

Approximate solution of KdV-Burgers equation using improved PINNs algorithm

Burgers' pinns with implicit euler transfer learning

Operator Learning Enhanced Physics-informed Neural Networks for Solving Partial Differential Equations Characterized by Sharp Solutions

Higher-order error estimates for physics-informed neural networks approximating the primitive equations

Wave simulation in non-smooth media by PINN with quadratic neural network and PML condition

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

VW-PINNs: A volume weighting method for PDE residuals in physics-informed neural networks

Error Analysis of Physics-Informed Neural Networks for Approximating Dynamic PDEs of Second Order in Time

Enhancing PINNs for solving PDEs via adaptive collocation point movement and adaptive loss weighting

Equation identification for fluid flows via physics-informed neural networks

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

General-Kindred Physics-Informed Neural Network to the Solutions of Singularly Perturbed Differential Equations

Robust error estimates of PINN in one-dimensional boundary value problems for linear elliptic equations

Physics-informed Neural Networks for Advection–diffusion–langmuir Adsorption Processes