Predicting plasma pressure profiles with Gaussian process and a neural network in KSTAR based on magnetic signals

As an equilibrium plasma is a force-balanced state, i.e., J×B = ∇p, in a tokamak, it is conceivable that magnetic signals, perhaps, can be used to infer the pressure profile. If such an inference can be reliably performed in real time, then this will greatly advance tokamak operations, especially for future fusion power plants where minimal number of diagnostics are only available. Thus, we have performed feasibility tests on predicting pressure profiles solely based on control magnetic signals by using a neural network. The neural network is trained with KSTAR pressure profiles which are obtained from Thomson Scattering (TS) and Charge Exchange System (CES) diagnostics where the Gaussian process is applied to the measured data. The neural network takes in-vessel coil currents, poloidal field current and plasma current as inputs and outputs the pressure profile. We present our preliminary results on our proposed scheme of predicting pressure profiles and discuss possibility of using the scheme for tokamak operations.