UNIVERSITY OF BUCHAREST
FACULTY OF PHYSICS

Guest
2024-11-22 2:09
 HOME     CONFERENCES     SEARCH            LOGIN     NEW USER     IMAGES   


Conference: Bucharest University Faculty of Physics 2024 Meeting


Section: Theoretical and Computational Physics, High-Energy Physics, Applied Mathematics


Title:
Analysis of turbulent transport in tokamak plasmas through test-particle simulations and neural network predictions


Authors:
Ligia-Maria POMÂRJANSCHI (1,2), Dragos Iustin PALADE (1,2)


Affiliation:
1) National Institute for Laser, Plasma and Radiation Physics

2) Faculty of Physics, University of Bucharest


E-mail
ligia.pomarjanschi@inflpr.ro dragos.palade@inflpr.ro


Keywords:
tokamak fusion plasma, turbulent transport, test-particle simulations, neural networks


Abstract:
In this work, we detail the methodology for using neural networks (NNs) to predict turbulent transport. The study focuses on magnetized fusion plasmas in tokamak devices, where the radial transport is driven through electrostatic microturbulence. This transport is quantified through macroscopic transport coefficients, i.e. convection and diffusion, which are obtained using test-particle simulations, or direct numerical simulations (DNS) [1]. At a statistical level, DNS simulates the motion of charged particle guiding centers in stochastically generated random fields associated with turbulence. Using the DNS numerical code, we created a database of plasma parameters and transport coefficients, on which we trained and tested the NN model [2]. Our results show that the NNs are orders-of-magnitude faster than traditional test-particle simulations, with a validation error below 2% and excellent agreement between predictions and real data [3]. Additionally, in terms of extrapolation and prediction, the neural network outperforms spline interpolation.


References:

[1] Palade, D.I., Vlad, M. Fast generation of Gaussian random fields for direct numerical simulations of stochastic transport. Stat Comput 31, 60 (2021).

[2] D.I. Palade, Predicting the turbulent transport of cosmic rays via neural networks. JCAP 2024(01), (2024), 002.

[3] L. M. Pomârjanschi, Neural networks for turbulent transport prediction in a simplified model of tokamak plasmas. 2024 Plasma Phys. Control. Fusion 66 065007