Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
P: Fachverband Plasmaphysik
P 1: Helmholtz Graduate School I - Theory
P 1.6: Vortrag
Montag, 5. März 2018, 12:35–13:00, A 0.112
Analysing the performance of neural networks on reconstructing edge plasma properties in Wendelstein 7-X — •Marko Blatzheim1,2, Daniel Böckenhoff1, Hauke Hölbe1, Thomas Sunn Pedersen1, and Roger Labahn2 — 1MPG IPP, Greifswald, Germany — 2University Rostock, Rostock, Germany
Artificial neural networks are a key technology to benefit from large amounts of data. The nuclear fusion experiment Wendelstein-7X is a fully optimized stellarator with the main goal to demonstrate steady state capability of fusion reactors. It is tried to analyse the edge plasma properties by artificial neural networks. Most data is based on simulations because experiment time is very limited. The same neural network should be able to deal with simulated results or experimental camera data as input. In a pre-processing step, characteristics of the simulation results and the camera images are extracted. These are expected to be sufficiently similar. The neural network performance for different such parametrizations is compared. Depending on the parametrization dimensionality, more complex neural network structures can be investigated. The most promising parametrizations will be used for more complicated plasma property reconstructions and predictions.