Erlangen 2018 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
P: Fachverband Plasmaphysik
P 5: Helmholtz Graduate School II
P 5.2: Vortrag
Montag, 5. März 2018, 16:40–17:05, A 0.112
Bayesian modelling of multiple diagnostics at Wendelstein 7-X using the Minerva framework — •Sehyun Kwak1,2, Jakob Svensson2, Sergey Bozhenkov2, Humberto Trimino Mora2, Udo Höfel2, Andrea Pavone2, Maciej Krychowiak2, Andreas Langenberg2, and Young-chul Ghim1 — 1Department of Nuclear and Quantum Engineering, KAIST, Daejeon 34141, Korea — 2Max-Planck-Institut für Plasmaphysik, 17491 Greifswald, Germany
Consistent inference of physics parameters and their uncertainties for large scale experiments requires the capability of handling the physics models of multiple sophisticated diagnostic systems. The Minerva framework has been developed for scientific inference and Bayesian modelling for complex systems, and is the standard analysis infrastructure for the W7-X experiment. It will be shown how Bayesian models implemented in the Minerva framework are capable of inferring electron temperature and density profiles from multiple diagnostic (Thomson scattering, interferometer, He-beam) data in a consistent way. The physics models for each diagnostic have been implemented and analysed individually as well as combined. The profiles are modelled by Gaussian processes with hyperparameters for varying length scales determined by a Bayes Occam*s razor criteria. The full posterior of profiles, hyperparameters, and calibration are explored by Markov chain Monte Carlo sampling. The results show all possible combinations of profiles, hyperparameters, and calibration with their associated uncertainties. Calibration of the Thomson scattering system is automatically handled by the combined model.