Feedforward shape control and neural net equilibrium modeling on NSTX-U

The NSTX-U shape control algorithm relies entirely on feedback control to track target shapes. Control performance can be improved by instead using feedback to adjust coil currents around a reference trajectory, i.e. feedback plus feedforward control. To this end, a design tool that translates target shape evolutions into coil current trajectories has been developed and validated with previous campaign data. Additionally, we report on development of several neural networks related to equilibrium modeling in NSTX-U. These networks can be used as standalone tools, or in conjunction with the feedforward coil currents planner to perform fast simulations. The networks include: (1) a neural network free-boundary solver that identifies equilibrium flux surfaces given the coil currents and internal profiles, (2) a convolutional neural network that performs the inverse problem, identifying coil currents from the flux surfaces, and (3) a neural net estimator of the nonrigid plasma response. The ground truth data for the nonrigid plasma response is obtained using the gspert code. This quantity is an estimate of how the plasma redistributes in response to external coil current perturbations or changes in Ip, Betap, and Li, and is important for shape control algorithm design.