Predictive JET current ramp-up modelling using QuaLiKiz-neural-network

This work applies the coupled JINTRAC and QuaLiKiz-neural-network (QLKNN) model on the ohmic current ramp-up phase of a JET D discharge. The chosen scenario exhibits a hollow T e profile attributed to core impurity accumulation, which is observed to worsen with the increasing fuel ion mass from D to T. A dynamic D simulation was validated, evolving j, n e, T e, T i, n Be, n Ni, and n W for 7.25 s along with self-consistent equilibrium calculations, and was consequently extended to simulate a pure T plasma in a predict-first exercise. The light impurity (Be) accounted for Z eff while the heavy impurities (Ni, W) accounted for P rad. This study reveals the role of transport on the T e hollowing, which originates from the isotope effect on the electron-ion energy exchange affecting T i. This exercise successfully affirmed isotopic trends from previous H experiments and provided engineering targets used to recreate the D q-profile in T experiments, demonstrating the potential of neural network surrogates for fast routine analysis and discharge design. However, discrepancies were found between the impurity transport behaviour of QuaLiKiz and QLKNN, which lead to notable T e hollowing differences. Further investigation into the turbulent component of heavy impurity transport is recommended.