Performance Analysis of the Neural Network Solution of Advection-Diffusion-Reaction Problem

Due to its capability on solving various problems, ANN has developed rapidly with many modifications, including the one with capability to solve physical systems without available external datasets. The so-called Physics Informed Neural Network (PINN) is implemented in this study to solve one-dimensional nonlinear finite boundary Advection-Diffusion-Reaction (ADR) problem. Different setups and cases are experimented to analyze its performance. The results are also compared with standard numerical solution. We show that the obtained result using proposed method is close to the existing solution with negligible absolute difference. We found that despite the disadvantages of slow computational time, PINN has better flexibility in terms of problem formulation and solution domain.