Real-time confinement mode classification and ELM onset prediction with the BES diagnostic system at DIII-D

The 2D Beam Emission Spectroscopy (BES) system at DIII-D can measure the localized pedestal dynamics of edge-localized mode (ELM) events and the edge turbulence dynamics associated with confinement regimes (L-mode, H-mode, QH-mode, and wide pedestal QH-mode). Here, we report on machine learning (ML) models for the real-time prediction of ELM onset and the real-time classification of the confinement regime using the 2D BES real-time data stream. The models will be deployed on a high-throughput FPGA accelerator for integration in the real-time plasma control system (PCS). To facilitate the avoidance or mitigation of impending ELM events by the real-time PCS, the BES ML models will generate a real-time output signal that corresponds to ELM onset likelihood. Similarly, to facilitate the access and sustainment of enhanced confinement regimes, the BES ML models will generate real-time output signals that correspond to confinement regime indicators. In addition, we report on a flexible feature space to support multiple simultaneous real-time tasks such as ELM onset prediction, confinement mode classification, and disruption prediction. Finally, we explore the feasibility to monitor in real-time the radial electric field (Er) shear and turbulent Reynolds stress in the pedestal.