Optimal Real-time Profile Regulation on EAST using Self-triggered Model Predictive Control
L. Yang, Z. Wang, S.T. Paruchuri, X. Song, E. Schuster
11th ITER International School on “ITER Plasma Scenarios and Control”
San Diego, California, USA, July 25-29, 2022
Actively controlling the safety factor profile, or just the q profile,
in tokamaks may provide MHD-stable plasma conditions and facilitate steady-state
operation. The term profile refers to the dependence of the plasma property,
in this case the safety factor, on the spatial coordinate on the poloidal
cross section from the magnetic axis to the plasma boundary. The q profile,
which provides a measurement of the pitch of the magnetic field confining
the plasma in tokamaks, is a function of the gradient of the magnetic
poloidal flux profile (Ψ-profile). Several control approaches have been studied
in the last decade to regulate and shape the q profile. The problem is
high dimensional due to the dependence of the controlled variable on space.
Moreover, the dynamics of the q profile is nonlinearly coupled with the
kinetic plasma conditions (temperature and density of the plasma). Due
to this complexity, an effective solution demands a model-based control
approach. A Model Predictive Control (MPC) strategy based on the magnetic
diffusion equation, which governs the dynamics of the poloidal magnetic
flux profile, is proposed in this work for the EAST tokamak. The finite-time
optimal control problem (FHOCP) is formulated by minimizing the tracking
error between actual and desired gradients of the poloidal magnetic flux
subject to both actuator and dynamic constraints. The solution of this
optimization problem in real time demands a simplified model based on a
linearized and spatially discretized version of the original model. As
a novelty with respect to previous MPC solutions, a state observer is
developed in this work to estimate the error between the original and
simplified models and to correct the predictions by the simplified model
within the MPC scheme. In addition, a self-triggered mechanism is implemented
within the FHOCP to prevent unnecessary computations arising in fixed
sampling-time MPC schemes.
*Supported by the US DOE under DE-SC0010537.