Recent progress on deep learning-based disruption prediction algorithm in HL-2A tokamak

Disruption prediction and mitigation is a crucial topic,especially for future large-scale tokamaks,due to disruption’sconcomitant harmful effects on the devices.On this topic,disruption prediction algorithm takes the responsibility to giveaccurate trigger signal in advance of disruptions,therefore the disruption mitigation system can effectively alleviate theharmful effects.In the past 5 years,a deep learning-based algorithm is developed in HL-2A tokamak.It reaches a truepositive rate of 92.2%,a false positive rate of 2.5%and a total accuracy of 96.1%.Further research is implementedon the basis of this algorithm to solve three key problems,i.e.,the algorithm’s interpretability,real-time capability andtransferability.For the interpretability,HL-2A’s algorithm gives saliency maps indicating the correlation between thealgorithm’s input and output by perturbation analysis.The distribution of correlations shows good coherence with thedisruption causes.For the transferability,a preliminary disruption predictor is successfully developed in HL-2M,a newlybuilt tokamak in China.Although only 44 shots are used as the training set of this algorithm,it gives reasonable outputswith the help of data from HL-2A and J-TEXT.For the real-time capacity,the algorithm is accelerated to deal with an inputslice within 0.3 ms with the help of some adjustments on it and TFLite framework.It is also implemented into the plasmacontrol system and gets an accuracy of 89.0%during online test.This paper gives a global perspective on these results anddiscusses the possible pathways to make HL-2A’s algorithm a more comprehensive solution for future tokamaks.

Recent Results on Field Reversed Configurations from the Translation, Confinement and Sustainment Experiment

The field-reversed configuration (FRC) offers an attractive alternative approach to magnetically confined fusion because of its extremely high β, simple linear geometry, and natural divertor for helium ash removal. Multi-hundred eV and high density FRCs have been produced using the standard Field Reversed Theta Pinch (RFTP) method, with a confinement scaling that leads to fusion conditions. These FRCs are, however, limited to only tens of mWb fluxes and sub-msec lifetime. Recent progress has been made in building up the flux and sustaining the FRC current using Rotating Magnetic Fields (RMF) in the Translation, Sustainment, and Confinement (TCS) facility at the University of Washington. TCS has demonstrated formation and steady-state sustainment of standard, flux-confined, prolate FRCs. The RMF also provides stability for the n = 2 rotational mode, which is the dominant global instability observed experimentally. Simple calculations show that a strong radially inward force imposed by the RMF increases proportionally to any local outward deformation of the plasma cross section. Evidence of this has been experimentally demonstrated, and the effects of various RMF antenna geometries studied. High temperature FRCs could also be produced in TCS by translating high energy plasmoids formed in the normal theta pinch manner into the confinement chamber containing the RMF antennas. Extremely interesting results were obtained for this translation and capture process. The plasmoids can survive the violent dynamics of supersonic reflections off magnetic mirror structures, producing a stable high-β, near-FRC state with substantial flux conversion from toroidal to poloidal. This is a tribute not only to the robustness of FRCs, but also to the tendency of an FRC to assume a preferred state for a magnetized plasma. The magnetic helicity, as inferred by a simple interpretive model, is approximately preserved, possibly conforming to a high-β relaxation principle.