Summary
This project develops machine learning tools for predicting turbulent transport in tokamak devices. The numerical foundation is the T3ST code, a highly parallelized Fortran solver designed to follow charged-particle dynamics in tokamak-like electromagnetic environments.
Turbulence is represented through synthetically generated random fields in field-aligned coordinates. This allows T3ST to compute transport coefficients through Lagrangian averaging while remaining rooted in a statistical description of turbulence.
The project uses T3ST to evaluate transport coefficients across scenarios defined by equilibrium, particle, and turbulence parameters. The resulting numerical database will support the training and validation of machine learning models for fast turbulent-transport prediction.
In the final phase, feed-forward neural networks and regression-based approaches will be used to provide fast surrogate predictions, complementing first-principles simulations in parameter regimes where direct computation is expensive.