Simulation Prediction of Heat Transport with Machine Learning in Tokamak Plasmas

Abstract

Machine learning opens up new possibilities for research of plasma confinement. Specifically, models constructed using machine learning algorithms may effectively simplify the simulation process. Previous first-principles simulations could provide physics-based transport information, but not fast enough for real-time applications or plasma control. To address this issue, this study proposes SExFC, a surrogate model of the Gyro-Landau Extended Fluid Code (ExFC). As an extended version of our previous model ExFC-NN, SExFC can capture more features of transport driven by the ion temperature gradient mode and trapped electron mode, using an extended database initially generated with ExFC simulations. In addition to predicting the dominant instability, radially averaged fluxes and radial profiles of fluxes, the well-trained SExFC may also be suitable for physics-based rapid predictions that can be considered in real-time plasma control systems in the future.