Development of COTSIM+NUBEAMNet One-Dimensional Predictive Simulation Capability for Scenario Optimization in NSTX-U
B. Leard, C. Clauser, H. Al Khawaldeh, T. Rafiq, E. Schuster
Division of Plasma Physics (DPP) Annual Meeting of the American Physical Society (APS)
Spokane, WA, USA, October 17-21, 2022
Plasma transport predictive models require a sufficiently small
computational effort to enable their use in model-based scenario-control
design. The Control Oriented Transport SIMulator (COTSIM), which captures
the control-relevant physics involved in a tokamak discharge, has been
developed to fulfill these computational-time constraints and to serve
the specific purpose of control design. In this work, the predictive
capabilities in COTSIM for NSTX-U scenarios have been significantly
improved. The dynamics of the plasma is modeled by the magnetic diffusion
equation, the heat transport equation for electrons and ions, and the
angular momentum equation. The current, torque, and heating depositions
by the neutral beam injectors are modeled using a neural network,
NUBEAMNet [1], which reproduces the results of NUBEAM in a fraction of
the computation time demanded by the original Monte Carlo code. Semi-empirical
transport models, Coppi-Tang and Bohm/gyro-Bohm, and fixed boundary
equilibrium solvers are also integrated with the transport equations.
Predictions by this fast-simulation capability are validated against
those by TRANSP. This new version of COTSIM for NSTX-U will enable a
myriad of plasma-control applications, including model-based scenario
optimization.
[1] M.D. Boyer et al 2019 Nucl. Fusion 59 056008.
*Supported by the US DOE under DE-SC0021385.