Realtime capable first principle based modelling of tokamak turbulent transport

Transport in the tokamak core is dominated by turbulence driven by plasma microinstabilities. When calculating turbulent fluxes, maintaining both a first-principle-based model and computational tractability is a strong constraint. We present a pathway to circumvent this constraint by emulating quasilinear gyrokinetic transport code output through a nonlinear regression using multilayer perceptron neural networks. This recovers the original code output, while accelerating the computing time by five orders of magnitude, allowing realtime applications. A proof-of-principle is presented based on the QuaLiKiz quasilinear transport model, using a training set of five input dimensions, relevant for ITG turbulence. The model is implemented in the RAPTOR real-time capable tokamak simulator, and simulates a 300s ITER discharge in 10s. Progress in generalizing the emulation to include 12 input dimensions is presented. This opens up new possibilities for interpretation of present-day experiments, scenario preparation and open-loop optimization, realtime controller design, realtime discharge supervision, and closed-loop trajectory optimization.