Fusion plasma turbulence simulation with neural network surrogate models

Turbulent transport in toroidal magnetic confinement devices, such as tokamaks, is one of the limiting factors for achieving viable fusion energy. Reactor design and plasma scenario optimisation demands both accurate and tractable predictive turbulence calculations. Neural network surrogate physics models provides a pathway to this goal. Databases of reduced order turbulence model output act as training sets for neural network regression. A key aspect is the customisation of output regression variables and optimisation cost functions in a physics-informed manner. The resultant surrogate model is 1012 times faster than direct numerical simulation and 106 times faster than the reduced order model itself.


About the speaker

Jonathan Citrin leads a computational physics research group at the Dutch Institute for Fundamental Energy Research (Differ). His focus is on plasma simulation for fusion energy. His specific emphasis is on the development and deployment of fast and accurate plasma turbulence models targeted for reactor design, optimisation, and control-oriented applications. This encompasses aspects of high-fidelity direct numerical simulation, reduced order modelling, and machine learning for fast surrogate model construction. A physicist by training, Citrin received his BSc from Tel Aviv University, MSc from the Weizmann Institute, and PhD from the Technical University of Eindhoven.