Real-time plasma confinement mode classification with deep neural networks and high-bandwidth edge fluctuation measurements in DIII-D

To demonstrate real-time confinement mode classification, we use the 2D beam emission spectroscopy (BES) diagnostic system [1] to capture localized density fluctuations in the pedestal region for long wavelength turbulent modes at a 1 MHz sampling rate. BES data was collected for ~150 DIII-D discharges in either L-mode, H-mode, QH-mode, or wide pedestal (WP) QH-mode to develop a deep-learning model for real-time confinement regime classification with the BES real-time data stream. These classification models can achieve accuracy and F₁ scores of 0.81 and 0.80, respectively. Moreover, our models are designed for and will be deployed on a high-throughput compute accelerator, such as a field programmable gate array (FPGA), for integration in the DIII-D plasma control system (PCS). This activity will demonstrate the feasibility for real-time data analysis of fluctuation diagnostics in future devices such as ITER, where potential risk of transient events is far greater and the need for reliable plasma performance is urgent.