A reduced-order model to estimate first wall particle and heat fluxes for systems codes

•development of TOKES surrogate models to quickly compute heat/ion far-SOL FW loads.•methodology uses PCA and novel information criteria expression for model evaluation.•identifies load profiles dependencies to filament temperature, density and velocity. Current systems codes (SCs) do not assess the...

Full description

Saved in:
Bibliographic Details
Published in:Fusion engineering and design Vol. 204; p. 114491
Main Authors: Pomella Lobo, Tiago, Pestchanyi, Sergey, Maione, Ivan Alessio
Format: Journal Article
Language:English
Published: Elsevier B.V 01-07-2024
Subjects:
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:•development of TOKES surrogate models to quickly compute heat/ion far-SOL FW loads.•methodology uses PCA and novel information criteria expression for model evaluation.•identifies load profiles dependencies to filament temperature, density and velocity. Current systems codes (SCs) do not assess the impact of plasma filament parameters and dynamics in the SOL when performing systems-level analyses. Simulation tools based on simplified turbulent transport models could address such need, such as the TOKES code, but these tend to run in timescales that are prohibitive for incorporation in SCs. This work presents a technique to build surrogate models for SCs, obtained with the modes of Reduced-Order Models (ROMs), and is exemplified with results from TOKES. These surrogates were developed with Principal Components Analysis (PCA), multivariate regression against TOKES inputs (filament temperatures, densities, and ejection speeds) and k-fold cross-validation, applied to the results after transformations based on rational powers of the inputs to optimize the prediction capabilities of the surrogates. Selection was performed with a modified Kullback-Leibler Divergence (KLD), and validation, with withheld cases. Main results include validation of the methodology due to the relatively low number of modes needed to represent more than 90% of the data variance, and the data transformation exponents that optimize the regression. [Display omitted]
ISSN:0920-3796
1873-7196
DOI:10.1016/j.fusengdes.2024.114491