Die DPG-Frühjahrstagung in Hannover musste abgesagt werden! Lesen Sie mehr ...

Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

P: Fachverband Plasmaphysik

P 5: Poster Session 1

P 5.12: Poster

Montag, 9. März 2020, 16:30–18:30, Empore Lichthof

Fast neural network surrogates for VMEC MHD equilibrium code — •Andrea Merlo, Daniel Böckenhoff, Thomas Sunn Pedersen, and The W7-X Team — Max-Planck-Institut für Plasmaphysik, Greifswald, Germany

MHD equilibrium codes, such as VMEC or PIES, are widely used among the fusion community for a variety of applications, ranging from MHD stability to reconstruction of plasma parameters. However, these codes usually require long computing times to converge, ranging from tens to hundreds of seconds per plasma configuration, preventing their adoption in real-time control scenarios or within extensive optimization runs. Data-driven methods for physical simulation codes allow to trade offline computation and memory footprint in favor for improved runtime performance. We are developing two machine-learning methods which act as fast surrogates for the VMEC code. These approaches comprise a Deep Neural Network (DNN) and a Physics-informed Neural Network (PiNN), where physical information regarding the system is provided via learning function. Additionally, to enable efficient training processes, our methods include subspace simulation techniques.