Bremen 2017 – wissenschaftliches Programm

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

P: Fachverband Plasmaphysik

P 15: Helmholtz Graduate School II

P 15.11: Poster

Dienstag, 14. März 2017, 16:30–18:30, HS Foyer

Inference of plasma parameters from an X-ray imaging diagnostic using neural networks — •Andrea Pavone, Jakob Svensson, and Andreas Langenberg — Max-Planck-Institute for Plasma Physics, Wendelsteinstraße 1, 17491 Greifswald, Germany

The W7X X-ray Imaging Crystal Spectroscopy system collects X-rays emitted in the interaction between electrons and ion impurities in the plasma. The light is collected along several lines of sight in the beam shaped poloidal cross-section of the torus. A forward model for this diagnostic is implemented in the Minerva framework, a Bayesian modelling framework which in this case allows the inference of plasma profiles via a nonlinear tomographic inversion of the measured images. This approach makes use of algorithms such as Markov Chain Monte Carlo (MCMC) and Maximum a Posteriori (MAP), which require the forward model to be run several time before a solution is found. This makes the data analysis relatively slow. Neural networks are an emerging class of algorithms for adaptive basis function regression. The regression is done by fitting the network model to a training set of images and corresponding profiles. Once the network is trained, it can be used on real data to provide real time analysis of measurements. This approach is easily generalizable to any diagnostic implemented in the Minerva framework. The neural network architecture will be described, together with the choices related to the creation of the training set, and results from the application to measured data. Finally, a comparison with the standard Bayesian inference strategy implemented within the Minerva framework is provided.