Uncertainty Quantification for Model Predictive Control of Tokamak Plasma Boundary Simulations with SOLPS-ITER

We deploy the SOLPS-ITER multi-fluid plasma and kinetic neutrals coupled transport solver to simulate dynamics of the tokamak plasma boundary in response to strategies for heat-flux mitigation on the divertor target. Several system identification approaches are implemented in the reduced modeling required for real-time predictive control. We merge Bayesian inference with the sparse identification of governing equations to develop a framework for model evaluation and selection. Analytic expressions for the steady-state correlations between upstream and downstream conditions are used as a reference for coupling scalar ODEs of the SOLPS-ITER computational domain. Training and testing over a series of gas puff actuation sequences for detachment control on the tokamak divertor are tuned to parameters for realistic integration with experimental devices. We demonstrate the successful application of adaptable and interpretable models in the feedback control of SOLPS-ITER simulations within the IPS integrated modeling environment.