Prediction of the Evolution of Tokamak Plasma Profiles Using Machine Learning

Temporal-evolution predictions of kinetic plasma profiles using data-driven approaches can be valuable in studying plasma transport mechanisms in tokamaks. The Grad-Shafranov (GS) equation describes the force balance in magnetohydrodynamic plasma equilibrium. A realistic plasma geometry’s GS equation can be numerically solved using the EFIT solver, which generally requires pressure and current density profiles as inputs. Typically, only magnetic measurements are used within the EFIT reconstruction and the results deviate from experimental profiles. More advanced reconstructions are constrained by the experimental measurements of the internal profiles and are known as kinetic equilibria. Automatic kinetic equilibrium reconstructions are being developed and can be used as inputs for plasma stability analysis, however, they do not consider temporal behavior. A data-driven approach was taken to predict the temporal-evolution of the plasma profiles. Deep learning methods are explored in the task of predicting temporal-evolutions of the kinetic plasma profiles with the potential to study the temporal-evolution of plasma transport in conjunction with transport models.