Machine-learning Algorithms Based on Time-series Data for Real-time Classification of Nuclear Fusion Plasmas in KSTAR

Abstract

A high confinement mode (H-mode)1) was selected as one of the basic operation scenarios of ITER.2) The H-mode is a phenomenon in which the plasma confinement performance suddenly increases because a transport barrier is formed near the plasma boundary. However, although there are advantages obtained in the H-mode, instabilities at the plasma boundary also appear, such as edge localized modes (ELMs). The problem with the ELMs is that the transport barrier periodically collapses (ELM crash). As a result, high temperature energy and particles are released, damaging the plasma-facing components. Therefore, various studies for controlling the ELMs are continuously conducted in various tokamak devices worldwide. The most effective method to prevent ELM crashes is resonant magnetic perturbation (RMP). Also, the most problematic phenomenon among various instability phenomena of plasma, including the ELMs, is plasma disruption.3) The disruption is a phenomenon in which the plasma inside the tokamak loses heat and electromagnetic energy very quickly and disappears. This sudden energy loss induces a considerable heat flux to the plasma-facing components. Furthermore, it applies a strong electromagnetic force to the surrounding conductor to damage the tokamak device, so it is an unstable phenomenon that must be resolved.