Application of Gaussian process regression to plasma turbulent transport model validation via integrated modelling

Abstract

This paper outlines an approach towards improved rigour in tokamak turbulence transport model validation within integrated modelling. Gaussian process regression (GPR) techniques were applied for profile fitting during the preparation of integrated modelling simulations allowing for rigourous sensitivity tests of prescribed initial and boundary conditions as both fit and derivative uncertainties are provided. This was demonstrated by a JETTO integrated modelling simulation of the JET ITER-like-wall H-mode baseline discharge #92436 with the QuaLiKiz quasilinear turbulent transport model, which is the subject of extrapolation towards a deuterium-tritium plasma. The simulation simultaneously evaluates the time evolution of heat, particle, and momentum fluxes over ∼ 10 confinement times, with a simulation boundary condition at ρ tor = 0.85. Routine inclusion of momentum transport prediction in multi-channel fluxdriven transport modelling is not standard and is facilitated here by recent developments within the Qua-LiKiz model. Excellent agreement was achieved between the fitted and simulated profiles for n e , T e , T i , and Ω tor within 2σ, but the simulation underpredicts the mid-radius T i and overpredicts the core n e and T e profiles for this discharge. Despite this, it was shown that this approach is capable of deriving reasonable inputs, including derivative quantities, to tokamak models from experimental data. Furthermore, multiple figures-of-merit were defined to quantitatively assess the agreement of integrated modelling predictions to experimental data within the GPR profile fitting framework.

Scrape-off layer transport and filament characteristics in high-density tokamak regimes

“…The ion saturation current and floating potential electronics have a bandwidth from 0.1 to 10 MHz with anti-aliasing filters at the Nyquist frequency: the acquisition frequency was set between 1 MHz to 5 MHz for different discharges used throughout the paper. The profiles in the SOL have been obtained combining the data from RCP with the ones from Thomson scattering diagnostic obtained in adjacent time instants: for TCV the profiles shown throughout the paper have been fitted using a Gaussian Process regression technique (details can be found in [28] together with the link to the available software tool). The method allows for a proper determination of the fit and corresponding errors, as well as of the density gradient with the corresponding errors.…”

Shaping effects on scrape-off layer plasma turbulence: A rigorous validation of three-dimensional simulations against TCV measurements

“…An overview of best practices and examples of validation in fusion is given in Refs. [19][20][21][22]. These examples reveal that the validation effort carried out by the fusion community has mostly focused on core turbulence.…”

Overview of the JET preparation for deuterium–tritium operation with the ITER like-wall

“…These simulations correspond to an averaged 500 ms time-window during discharge flattop. A Gaussian Process Regression fit is performed on the kinetic profile data, and the distribution average is used as initial condition 13 . The current, temperature and density profiles are then evolved over multiple energy confinement times until the temperature and density profiles are in stationary-state, and compared to the experimental fits.…”

“…As QLKNN is only applicable for turbulent transport in the tokamak core, we evolve temperature and density only inside ρ N,tor = 0.85, and include a proxy transport coefficient for sawtooth-induced transport in the deep core for all simulations in this work. 13 . Appendix A contains a full overview of the applied settings.…”

“…Future work will improve the QLKNN model by extending to larger input space, focusing on the impurity density gradient, and multiple-ion transport important for multiple-isotope fuelling applications and impurity transport. Additionally, using a robust method to fit a large amount of experimental kinetic profiles 13 , one can base a training set on experimental data, instead of the hyperrectangle methodology described here. This allows for more input dimensions to be used, as well as including rotation by using the native QuaLiKiz rotation model, instead of a rotation rule as described here.…”