Equilibrium reconstruction of axisymmetric plasmas by combining Gaussian process regression and Markov chain Monte Carlo sampling
- 29 November 2024
- journal article
- research article
- Published by IOP Publishing in Plasma Physics and Controlled Fusion
- Vol. 67 (1) , 015006
Abstract
Reliable equilibrium reconstruction is indispensable for understanding and controlling hot magnetized plasmas to achieve fusion reactors. In axisymmetric systems, current and pressure profiles that satisfy the force balance conditions are given by the Grad–Shafranov (GS) equation. While many novel approaches have been developed to swiftly and robustly find an optimum solution of the GS equation, approaches based on a single solution search may not be adaptable if diagnostics fail to provide sufficient constraints. Here, we investigate the solution space of the GS equation when only basic edge magnetic measurements are available. By combining Gaussian process regression and Markov chain Monte Carlo sampling within the Bayesian framework, we treat each current element as an independent variable and evaluate the probability distribution that describes all possible solutions. We have applied this inference frame to the geometry of the PLATO tokamak and shown that the flux surface locations can be determined relatively well only from 16 pick-up coils, 4 flux loops and a diamagnetic loop. On the other hand, the toroidal current density is inferred with limited success, and the inferences of the safety factor and pressure profiles are difficult. The characterization of possible choices of equilibria realized by this inference framework will help optimize diagnostic setups for equilibrium reconstruction.Keywords
Funding Information
- Research Center for Mathematical and Data Science
- Kyushu University
- JSPS (23K13085)
- PRESTO
- JST
This publication has 23 references indexed in Scilit:
- GS-DeepNet: mastering tokamak plasma equilibria with deep neural networks and the Grad–Shafranov equationScientific Reports, 2023
- Fast equilibrium reconstruction by deep learning on EAST tokamakAIP Advances, 2023
- The measurements by diamagnetic loops in EASTFusion Engineering and Design, 2022
- Plasma parameter profile inference from limited data utilizing second-order derivative priors and physic-based constraintsPhysics of Plasmas, 2021
- Estimation and Uncertainties of Profiles and Equilibria for Fusion Modeling CodesFusion Science and Technology, 2020
- Past and present experiments toward PLATO projectAIP Conference Proceedings, 2018
- Inference of field reversed configuration topology and dynamics during Alfvenic transientsNature Communications, 2018
- Inferring time derivatives including cell growth rates using Gaussian processesNature Communications, 2016
- Bayesian soft X-ray tomography using non-stationary Gaussian ProcessesReview of Scientific Instruments, 2013
- Reconstruction of current profile parameters and plasma shapes in tokamaksNuclear Fusion, 1985