Machine learning autoencoder models for compressing and reconstructing 2D BES data for real-time classification tasks.

Multi-channel fluctuation diagnostics capture the plasma dynamics. Here, we report on an effort to develop machine learning (ML) models for the real-time identification of edge-localized-mode (ELM) events and the turbulence properties of confinement regimes using the 2D Beam Emission Spectroscopy (BES) system at DIII-D. The "edge ML" models will be deployed on a high-throughput FPGA accelerator for integration in the real-time plasma control system (PCS). The models will generate reduced signals that correspond to ELM activity and turbulence dynamics, and the real-time PCS will be trained to avoid ELM regimes and to maintain advanced confinement regimes such as the wide pedestal QH-mode. The 2D BES system captures density perturbations imprinted in neutral beam emission at a 1 MHz frame rate. Here, we report on autoencoder neural networks to compress the spatial-temporal information in a low-dimension space. Using such an autoencoder, we plan to compress BES data for ELM classification and other classification tasks. Currently, we experiment with the number of hidden layers and network architecture to maximize the capability of the network to compress and reconstruct 2D BES data as measured by a mean squared error loss function.