SMuK 2021 – wissenschaftliches Programm

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

P: Fachverband Plasmaphysik

P 17: Poster II

P 17.23: Poster

Freitag, 3. September 2021, 14:00–16:00, P

Gaussian Process Surrogate Models for Uncertainty Quantification in Multiscale Turbulent Transport Simulations — •Yehor Yudin, Jalal Lakhlili, Onnie Luk, Udo von Toussaint, and David Coster — Max Planck Institute for Plasma Physics, Boltzmannstrasse2, 85748 Garching, Germany

One of the challenges in understanding fusion plasmas is quantifying the effects of micro-scale turbulent dynamics on energy and particle transport processes in a fusion device. In order to analyze such effects, one should numerically solve a model which couples system evolution on disparate spatial and temporal scales, as well as consider both aleatoric and epistemic uncertainty of such model. For such a solution the largest share of computational expense is spent on resolving turbulence related scales. This work proposes an application of a surrogate modelling approach to reduce computational costs for a solution in a case close to a quasi-steady state when it is sufficient to capture only statistics of turbulent dynamics. We studied a Multiscale Fusion Workflow that couples gyrofluid turbulence code GEM in flux tube approximation with core transport code ETS, and calculates transport coefficients from turbulent energy and particle fluxes. For that, we applied the VECMA toolkit to perform uncertainty quantification, as well as to train, test and utilize surrogate models. In this work, a data-driven probabilistic surrogate model based on Gaussian Process Regression is used to infer flux values computed by a turbulence code for given core profiles, and to calculate related uncertainties.