Use of Machine Learning Techniques to Evaluate TORIC Generated Profiles on NSTX in the start-up/ramp-up regime

The non-inductive start-up of a tokamak is a major challenge currently being addressed in the field of plasma physics. One method of heating the plasma during this process is the use of Fast Waves (FW) in the ion cyclotron range of frequency (ICRF). At the NSTX-U tokamak, the Radio Frequency (RF) current drive is achieved through the use of a high harmonic fast wave (HHFW) system consisting of a 12-strap antenna connected to six decoupled 30 MHz sources, which can provide up to 6 MW of RF power.

In this work, HHFW simulations in NSTX are carried out using the version of the full wave TORIC code able to treat the higher cyclotron harmonics [M. Brambilla, Plasma Phys. Control. Fusion 44 (2002) 2423]. The TORIC code provides the wave electric field, the ion/electron power absorption, and the electron current drive under various conditions such as, magnetic equilibrium, plasma composition, kinetic profiles, and toroidal wave number. These simulations are crucial for understanding the performance of the HHFW in NSTX and for optimizing its operation.

This work aims to compare the accuracy of Machine Learning (ML) techniques in reconstructing the current drive and power electron density profiles generated by the TORIC code. This will provide insight into the potential of ML methods to accurately recreate these output signals, which are important for understanding and optimizing the performance of the HHFW source.