Machine learning for core transport model validation

Pablo Rodriguez Fernanadez

PSFC

Tuesday, September 19, 2017

5:00pm

NW17-218

PSFC Student Seminars

This talk presents a machine learning framework that is developed for the validation of transport models in fusion research. A surrogate-based strategy using Gaussian processes and sequential parameter updates is used to achieve the combination of plasma input parameters that matches experimental heat fluxes and transport features simultaneously within error bars. Turbulence measurements and dynamic transport characteristics are used, for the first time, during the validation process of the TGLF reduced transport model. First results indicate that these machine learning algorithms are very suitable as a self-consistent and comprehensive validation methodology for plasma and fluid transport codes. Future work will focus on validating reduced transport models, such as TGLF, with integrated “whole-device” modeling frameworks to study transient phenomena in tokamak plasmas.