Neural network surrogate of QuaLiKiz using JET experimental data to populate training space

Within integrated tokamak plasma modeling, turbulent transport codes are typically the computational bottleneck limiting their routine use outside of post-discharge analysis. Neural network (NN) surrogates have been used to accelerate these calculations while retaining desired accuracy physics-based models. This paper extends a previous NN model, known as QLKNN-hyper-10D, by incorporating impact impurities, rotation, and magnetic equilibrium effects. is achieved adding light impurity fractional density ( n imp , / e) its normalized gradient, pressure gradient (?), toroidal Mach number M tor), flow velocity gradient. The input space was sampled based on experimental data from JET avoid curse dimensionality. resulting networks, named QLKNN-jetexp-15D, show good agreement with original QuaLiKiz both comparing individual quantity predictions within JINTRAC. profile-averaged RMS modeling simulations <10% for each five scenarios tested. non-trivial given potential numerical instabilities present highly nonlinear system equations governing transport, especially considering novel addition momentum flux model proposed here. An evaluation all 25 output quantities at one radial location takes ?0.1 ms, 104 times faster than model. JINTRAC tests performed in this study, using QLKNN-jetexp-15D resulted speed increase only 60–100 other physics modules become bottleneck.