Author: Yaduo LI.
Advisors: Joseph MORLIER.
ISAE SUPAERO, 2019.
This project aims to presnet the comparison results of different mythologies for the parametrized PDE problem. The test is firstly taken on a Steady-Advection-Diffusion problem. The following methods are included:
- Physics-Informed Neural Networks (PINN).
- Reduced Order Models using Proper Orthogonal Decomposition (POD) method. The reduced base is computed using a greedy algorithm and also the TensorFlow package.
- Gaussian Process to esitmate directly the computation results form a group of parameters.
- Kriging Model using the Partial Least Squares method (KPLS)
To luanch the testing notebook, the following should be used:
- Python version 3.7
- The needed packages:
- pyDOE and Scikit learn
- deepxde for the PINN method
- TensorFlow for the TensorFlow based POD method
- SMT for the KPLS method
- ipywidgets for the visualization of the results