Model predictive control of boundary plasmas using reduced models derived from SOLPS-ITER

Time-dependent simulations of the tokamak boundary plasma are performed using the SOLPS-ITER transport code to develop reduced models and model-predictive control (MPC) of the upstream and divertor conditions actuated by main ion and impurity gas puff. The reduced models are based on DMD and SINDy methods, which are data-driven algorithms that extract dynamic behavior to describe the underlying physical system. With DMD, the time evolution is described by discrete operators, while SINDy results in a sparse set of coupled ordinary differential equations. In either case, the model is used to predict the evolution of the current plasma state over a rolling time horizon and determine an optimal actuation sequence to best produce a target trajectory, subject to constraints. Feed-forward MPC actuation sequences input to SOLPS using a DIII-D configuration have been found to agree well with a target density trajectory. When prediction error exceeds a prescribed threshold the model can be updated using past data over a set time window. The MPC method is being implemented into an efficient module that can be called from SOLPS as an online controller. The reduced models are also compared to analytic results and data-mined correlations extracted from simulation and experimental data.