Preemptive RMP-driven ELM crash suppression automated by a real-time machine-learning classifier in KSTAR

Abstract

Suppression or mitigation of edge-localized mode (ELM) crashes is necessary for ITER. The strategy to suppress all the ELM crashes by the resonant magnetic perturbation (RMP) should be applied as soon as the first low-to-high confinement (L-H) transition occurs. A control algorithm based on real-time machine learning (ML) enables such an approach: it classifies the H-mode transition and the ELMy phase in real-time and automatically applies the preemptive RMP. This paper reports the algorithm design, which is now implemented in the KSTAR plasma-control system, and the corresponding experimental demonstration of typical high-δ KSTAR H-mode plasmas. As a result, all initial ELM crashes are suppressed with an acceptable safety factor at the edge (q95) and with RMP field adjustment. Moreover, the ML-driven ELM-crash-suppression discharges remain stable without further degradation due to the regularization of the plasma pedestal.

Integrated RMP-based ELM-crash-control process for plasma performance enhancement during ELM crash suppression in KSTAR

“…The ML-classifier-based RMP onset, originally designed for the pre-emptive ELM crash control, offers an advantage in enhancing β N compared to the conventional pre-set RMP onset. This β N enhancement in the pre-emptive RMP onset is primarily due to higher ion temperature in the plasma core region relative to that in the conventional RMP onset [37]. A turbulence and transport analysis in progress will provide a physics understanding of the increased core T i in the RMP onset coincident with the L-H transition.…”

“…However, for an efficient and stable β N -enhanced suppression phase, it is necessary to introduce the latest achievements related to the RMP technique. The pre-emptive RMP onset based on the real-time Machine Learning (ML) classifier [36], which automatically triggers RMP before the first ELM after the L-H transition, can obtain a higher ion temperature at the plasma core region compared to the conventional pre-set RMP onset [37]. The interactive I RMP control by the adaptive feedback RMP ELM controller balances β N enhancement and ELM crash suppression by optimizing I RMP [15,38,39], in contrast to the conventional pre-set I RMP control.…”

Highest fusion performance without harmful edge energy bursts in tokamak

“…However, avoiding extensive ELMs between the LH transition and the first ELM suppression is vital. Previous research has demonstrated that early RMP-ramp up 60,61 before the first ELM reduces ELMs during the early H-mode phase. Nevertheless, this approach often faced limitations due to uncertainties in determining the required conditions, including initial RMP amplitude for suppressing the first ELMs.…”

Tokamak edge localized mode onset prediction with deep neural network and pedestal turbulence

“…Here, with figure 4(c), we compare the network results with the L-H transition time (t LH ) in order to show that predicting the first ELM onsets is more optimal to perform ELM suppression tools than detecting the L-H transition. Due to the obvious fact that the first ELMs occur after the L-H transition, a way of detecting t LH based on a neural network with the series of D α and n e signals [23,33] has been made to activate ELM suppression controls from the beginning of H-modes, while avoiding the mode-locking in L-modes. Conversely, figure 4(c) shows that t LH apparently cannot be considered as an optimal indicator to respond to the first ELM onset (t fe ) as ∆t = t fe − t LH varies from discharge to discharge.…”

“…This RMP-driven operation can respond to the onsets of the ELMs in real time through optimizing the threedimensional coil currents [20][21][22]; however, ELM onset prediction or at least ELM detection methods [22] should be required. Also, this operation is typically programmed to be activated immediately after bifurcating from the low confinement mode (L-mode) as a pre-emptive way [21,23] in order to prevent the mode-locking [23] as well as suppress any possible ELMs. This pre-emptive control is due to a lack of predictive knowledge about when the first ELM will be developed after the H-mode transition.…”