Single Gaussian process method for arbitrary tokamak regimes with a statistical analysis

Abstract

Gaussian Process Regression (GPR) is a Bayesian method for inferring profiles based on input data. The technique is increasing in popularity in the fusion community due to its many advantages over traditional fitting techniques including intrinsic uncertainty quantification and robustness to over-fitting. Most fusion researchers to date have utilized a different GPR kernel for each tokamak regime. This requires a Machine Learning (or simpler) method to first predict the regime, choose the right kernel for that regime, and then use that kernel. The disadvantage of this method is that it requires an additional step, and it is unclear how well it will behave if the plasma enters a new, unexpected regime. We summarize our work developing a general kernel for all regimes (including radially-varying hyperparameters), utilizing heavy-tailed likelihood distributions to automatically handle data outliers, and using GPflow for full Bayesian inference via Markov chain Monte Carlo to sample hyperparameter distributions. We present a single GPR method that is robust across

A Gaussian process guide for signal regression in magnetic fusion

“…In [20] GPs are used for determining properties of micro-tearing modes in spherical tokamaks, while in [21] GPs are used for inferring axisymmetric plasma equilibrium from magnetic field and plasma pressure measurements and in [22] Bayesian inference is used to resolve Z eff profiles from line integrated bremsstrahlung spectra. In [3,[23][24][25] GPRs are used to fit experimental core temperature and density profiles, to infer second order derivative information from these profiles, and to propagate uncertainties, while in [26] GPs are explored for use in quasi-coherent noise suppression. At JET, GPs have been used to infer electron cyclotron emission spectra [27], and high resolution Thomson scattering and far infrared interferometer data in [28], while work in [29] is focused on using GPRs to quantify edge plasma evolution from experimental data.…”

Real‐time estimation of the electron temperature profile in DIII‐Dby leveragingneural‐networksurrogate models

“…[ 3 ] Both of these approaches are dependent on the assumption of the shape of the fitting function. Gaussian process regression is a statistical tool that is also being explored for the purpose of profile fitting, [ 4 ] but there is currently no widely available tool that uses this method for DIII‐D. All of these approaches are heavily reliant on just the diagnostic data and do not take into account much of the known physics of how the profile behaves.…”

Gaussian process-based quasi-coherent noise suppression in magnetic confinement devices with superconductors

“…Note that the GP achieves non-parametric regression using a multivariate normal distribution and a covariance kernel function taking into account of correlations determined by the data distance. The GP is a widely adopted non-parametric model in the fusion community when a parametric model describing measured signals is not suitable or limited such as plasma profile fittings [22][23][24][25][26] or tomographic reconstructions [27][28][29][30].…”