Gaussian Process Regression for Equilibrium Reconstruction in DIII-D and ITER Plasmas

Gaussian Process Regression is a Bayesian method for inferring profiles based on observed experimental data and physics constraints as probability input. The technique is increasing in popularity in the fusion community for its many advantages over traditional profile fitting techniques. Here, we present our progress on applying GPR techniques to Thomson scattering synthetic data [1], fitting experimental data within the EFIT reconstruction workflow, and contrasting with similar efforts within the fusion community. These applications include predictions of profiles as well as important quantities for equilibrium reconstruction such a pedestal location, height, and width, and the errors on these. We present our library, Unbaffeld, which provides open-source access to these GPR tools for the community both directly and through OMFIT. With new tokamaks running in novel regimes, GPR is more desirable than parameterized fitting due to the minimal required assumptions and lack of human intervention. To demonstrate this ability, we will perform a statistical analysis of GPR as applied to DIII-D data. We summarize the results and discuss how to extend it to full kinetic reconstructions for applications to ITER and future burning plasma devices such as FPP.