Preemptive ECCD stabilization of NTMs through machine learning on DIII-D

Neoclassical tearing modes (NTMs) are a resistive instability that pose a large risk not only to tokamak performance but can lead to violent disruptions that can damage hardware. NTM control techniques are considered essential by ITER and one proposed control scheme involves continuous preemptive electron cyclotron current drive (ECCD) at rational surfaces to replace the missing bootstrap current. However, while this approach has been demonstrated on current machines, it is restrictive and will use more heating power than is necessary for NTM control. To improve upon the basic control scheme, we present a novel ML-based NTM predictor model that provides an NTM probability metric with large warning times prior to NTM onset. Integrating this NTM survival metric into an ITER-like gyrotron control framework allows for multitasking gyrotrons to balance two control tasks: ECCD for active NTM suppression and broad off-axis ECCD to achieve an advanced, high-performance scenario. The controller actively steers the ECCD deposition to rational surfaces only when an NTM probability is high to save gyrotron power for other tasks. This multi-objective control was successfully demonstrated in experiment by maintaining the desired advanced non-inductive scenario while preemptively suppressing NTMs. This approach provides an intelligent scheme for allocation of gyrotron resources for both DIII-D's future planned gyrotron expansions and ITER's eventual electron-cyclotron dominant heated plasmas.