Investigation of neoclassical transport in an axisymmetric tokamak using a Neural Network

Toroidal geometry and non-uniform magnetic fields give rise to a class of heat and particle transport known as "neoclassical transport." Neoclassical transport can play an important role in magnetic fusion devices, especially in the presence of non-axisymmetric fields or in regions where turbulence is suppressed, where it can become a dominant mechanism for heat and particle loss in the plasma. Several codes exist that can calculate neoclassical transport, including NEO. Although these codes are relatively efficient, they are too slow to be used effectively inside optimization loops, for real-time control, or within large-scale fluid models, especially when multiple ion species are present or in non-axisymmetric magnetic fields. This project seeks to develop a fast method for evaluating neoclassical transport using machine learning, by training a neural network using the output of NEO calculations. This neural network will be able to accelerate neoclassical transport modeling. Initial results are calculated using a single ion species in axisymmetric Miller geometry. This work can be extended in the future to consider multiple ion species and non-axisymmetric geometries.