Search

Using machine learning (ML) techniques to develop disruption predictors is an effective way to avoid or mitigate the disruption in a large-scale tokamak. The recent ML-based disruption predictors have made great progress regarding accuracy, but most of them have not achieved acceptable cross-machine performance. Before we develop a cross-machine predictor, it is very important to investigate the method of developing a cross-tokamak ML-based disruption prediction model. To ascertain the elements which impact the model’s performance and achieve a deep understanding of the predictor, multiple models are trained using data from two different tokamaks, J-TEXT and HL-2A, based on an implementation of the gradient-boosted decision trees algorithm called LightGBM, which can provide detailed information about the model and input features. The predictor models are not only built and tested for performance, but also analyzed from a feature importance perspective as well as for model performance variation. The relative feature importance ranking of two tokamaks is caused by differences in disruption types between different tokamaks. The result of two models with seven inputs showed that common diagnostics is very important in building a cross-machine predictor. This provided a strategy for selecting diagnostics and shots data for developing cross-machine predictors.

A large number of runaway electrons (REs) with energies as high as several tens of mega-electron volt (MeV) may be generated during disruptions on a large-scale tokamak. The kinetic energy carried by REs is eventually deposited on the plasma-facing components, causing damage and posing a threat on the operation of the tokamak. The remaining magnetic energy following a thermal quench is significant on a large-scale tokamak. The conversion of magnetic energy to runaway kinetic energy will increase the threat of runaway electrons on the first wall. The magnetic energy dissipated inside the vacuum vessel (VV) equals the decrease of initial magnetic energy inside the VV plus the magnetic energy flowing into the VV during a disruption. Based on the estimated magnetic energy, the evolution of magnetic-kinetic energy conversion are analyzed through three periods in disruptions with a runaway current plateau.

The response of plasma toroidal rotation to the external resonant magnetic perturbations (RMP) has been investigated in Joint Texas Experimental Tokamak (J-TEXT) ohmic heating plasmas. For the J-TEXT’s plasmas without the application of RMP, the core toroidal rotation is in the counter-current direction while the edge rotation is near zero or slightly in the co-current direction. Both static RMP experiments and rotating RMP experiments have been applied to investigate the plasma toroidal rotation. The core toroidal rotation decreases to lower level with static RMP. At the same time, the edge rotation can spin to more than 20 km s−1 in co-current direction. On the other hand, the core plasma rotation can be slowed down or be accelerated with the rotating RMP. When the rotating RMP frequency is higher than mode frequency, the plasma rotation can be accelerated to the rotating RMP frequency. The plasma confinement is improved with high frequency rotating RMP. The plasma rotation is decelerated to the rotating RMP frequency when the rotating RMP frequency is lower than the mode frequency. The plasma confinement also degrades with low frequency rotating RMP.

The suppression of disruption-generated runaway electrons (REs) by supersonic molecular beam injection (SMBI) has been investigated on the J-TEXT tokamak. Experimental results demonstrate that the hydrogen injected by SMBI during plasma current flattop phase can provoke magnetic perturbations, which increase RE losses rapidly. The effective radial diffusion coefficient of REs due to SMBI is estimated as D r ≈ 16 m2 s−1. Based on this benefit, the SMBI has been used to explore the suppression of disruption-generated REs. In J-TEXT, RE current is created with rapid argon injection by a massive gas injection valve. It is found that hydrogen SMBI before disruption efficiently suppresses the generation of RE current.

Disruption mitigation is essential for the next generation of tokamaks. The prediction of plasma disruption is the key to disruption mitigation. A neural network combining eight input signals has been developed to predict the density limit disruptions on the J-TEXT tokamak. An optimized training method has been proposed which has improved the prediction performance. The network obtained has been tested on 64 disruption shots and 205 non-disruption shots. A successful alarm rate of 82.8% with a false alarm rate of 12.3% can be achieved at 4.8 ms prior to the current spike of the disruption. It indicates that more physical parameters than the current physical scaling should be considered for predicting the density limit. It was also found that the critical density for disruption can be predicted several tens of milliseconds in advance in most cases. Furthermore, if the network is used for real-time density feedback control, more than 95% of the density limit disruptions can be avoided by setting a proper threshold.

The plasma response to externally applied perturbation fields is investigated on the J-TEXT tokamak using a set of static resonant magnetic perturbation (SRMP) coils. Several different experimental results are obtained including partial or complete suppression of the existing m/n = 2/1 tearing mode, mode locking or non-uniform magnetohydrodynamic (MHD) oscillations. These results depend sensitively on the tearing mode frequency and the amplitude of the perturbation field. It is found that mode locking is most likely to happen at a lower rotation frequency (< ~5 kHz) and the threshold for mode locking has a linear relation with MHD frequency. However, complete suppression of the tearing mode happens in a region where the MHD frequencies are higher (~6 kHz). The experimental observations are explained by numerical simulations based on reduced MHD equations. The error field contributes to an offset between the mode-locking thresholds for the two opposite spatial phases of the SRMP, through which the intrinsic error field of J-TEXT can be estimated.

The high fraction of runaway current conversion following disruptions has an important effect on the first wall for next-generation tokamaks. Because of the potentially severe consequences of a large full current runaway beam on the first wall in an unmitigated disruption, runaway suppression is given a high priority. The behavior of runaway currents both in spontaneous disruptions and in D2 massive gas injection (MGI) shutdown experiments is investigated in the KSTAR tokamak. The experiments in KSTAR show that the toroidal magnetic field threshold, BT >2 T, for runaway generation is not absolute. A high fraction of runaway current conversion following spontaneous disruptions is observed at a much lower toroidal magnetic field of BT = 1.3 T. A dedicated fast valve for high-pressure gas injection with 39.7 bar is developed for the study of disruptions. A study of runaway current parameters shows that the conversion efficiency of pre-disruptive plasma currents into runaway current can reach over 80% both in spontaneous disruptions and in D2 MGI shutdown experiments in KSTAR.