Team, Visitors, External Collaborators
Overall Objectives
Research Program
Application Domains
Highlights of the Year
New Software and Platforms
New Results
Bilateral Contracts and Grants with Industry
Partnerships and Cooperations
XML PDF e-pub
PDF e-Pub

Section: New Results

Machine learning techniques for reduced model and stabilization

Participants: E. Franck, L. Navoret, V. Vigon (IRMA Strasbourg).

Just recently, we have begun to work on applications of machine learning techniques for the plasma simulation. This preliminary work is in the context of "Action exploratoire MALESI" and will really begin in 2020. The first point is the construction of a new closure for the fluid models using kinetic simulation as data. We have constructed 1D solvers for the Vlasov-Poisson equation with collisional operator and Compressible Navier-Stokes Poisson models. Comparing the models we observe that the classical Navier-Stokes closure is not sufficient when the Knudsen number is larger than 0.3-0.4. Currently we generate data using the Vlasov-Poisson code to train a neural-network for the closure. The second point is about the stabilization of the numerical method using CNN. We began a study to construct a Convulational Neural Network (CNN) to detect the Gibbs oscillations in the fluid simulations.