Towards real-time detachment control using reduced models derived from the SOLPS-ITER code

Time-dependent SOLPS-ITER simulations are used to develop reduced models for real-time control of the boundary plasma. This approach extends beyond typical linear control used on present-day tokamaks, offering the potential to respond to rapid transient behavior in addition to avoiding time-consuming tuning procedures. SOLPS-ITER is run to determine the plasma response to fuel ion and impurity seeding gas inputs. The full 2D simulation output is reduced to a set of experimentally measurable upstream and downstream quantities with the implementation of synthetic diagnostics, to test control of fully observable and limited access systems. Data-driven methods that identify sparse interpretable (non-black box) dynamic models are applied, including DMD, SINDy, and a novel kernel-based method to reduce sensitivity to noise. The system nonlinearity is stronger for impurity seeding as compared to fuel ion puffing, as well as when the divertor plasma approaches detached conditions, highlighting the need for model-based control. Model-predictive feedback control with adaptive retraining has been demonstrated and used to virtually control a DIII-D discharge and has also been recently implemented for simulating control of a hydrogen ITER plasma.