Data-Driven Control for Radiative Collapse Avoidance in Large Helical Device

Abstract

A radiative collapse predictor has been developed using a machine-learning model with high-density plasma experiments in the Large Helical Device (LHD). The model is based on the collapse likelihood, which is quantified by the parameters selected by the sparse modeling, including n_e, CIV, OV, and T_e,edge. The control system implementing this model has been constructed with a single-board computer to apply this predictor model to the LHD experiment. The controller calculates the collapse likelihood and regulates gas-puff fueling and boosts electron cyclotron resonance heating in real-time. In density ramp-up experiments with hydrogen plasma, high-density plasma has been maintained by the control system while avoiding radiative collapse. This result has shown that the predictor based on the collapse likelihood has the capability to predict a radiative collapse in real-time.