Erlangen 2018 – scientific programme
Parts | Days | Selection | Search | Updates | Downloads | Help
P: Fachverband Plasmaphysik
P 11: Helmholtz Graduate School - Poster
P 11.20: Poster
Tuesday, March 6, 2018, 16:15–18:15, Redoutensaal
Minerva neural network based surrogate models for real time inference of ion temperature profiles at Wendelstein 7-X — •Andrea Pavone1, Jakob Svensson1, Andreas Langenberg1, Novimir Pablant2, Robert C. Wolf1, and W7-X team1 — 1Max-Planck-Institut für Plasmaphysik, Teilinstitut Greifswald, D-17491 Greifswald, Germany — 2Princeton Plasma Physics Laboratory, 08540 Princeton, NJ, US
At the Wendelstein 7-X stellarator, tens of diagnostics produce a massive amount of data relevant to the measurement of plasma parameters. Forward models of many diagnostics are implemented in the Minerva framework where Bayesian inference is applied to complex systems constituting multiple diagnostics and physics models. Conventional inversion routines can take hours for a single data point, but Artificial Neural Networks (ANNs) can reduce the time by several orders of magnitude, making real-time analysis possible and providing a reliable alternative to such routines. We have trained ANNs exclusively on data synthesized within the Minerva framework, so that the trained ANNs constitute a surrogate model of the full physics model. We also focused on the quantitative estimation of the uncertainties of the ANN*s output, based on the stochastic properties of the optimization and the limited parameter space coverage of the training set with respect to novel measured data points. Results from applying ANNs to ion temperature profile inference from 2D X-ray spectral measurements will be compared with standard inversion routines, showing robustness and reliability for real time plasma parameter inference.