Classifying Alfvén eigenmodes using deep learning and CO₂ interferometer data at DIII-D

Deep learning models are trained using CO₂ interferometer data and human labels to detect the presence of Alfvén eigenmodes (AE) in 1112 discharges at DIII-D. The goal of this project is to demonstrate the potential of using deep learning models to accurately detect AE modes observed in tokamak plasmas. Resonant fast ions can drive AE modes unstable and degrade the plasma performance or energy confinement. Mode activity can be detected in the crosspower spectra of the CO₂ interferometer diagnostic. This task is time intensive and requires extensive domain knowledge. However, implementing machine learning models into the analysis routine would accelerate the process and increase the frequency of accurate predictions. Recent work produced a database of the occurrence of EAE, TAE, RSAE, BAE, LFM, and EGAM activity [Heidbrink, et al., NF ‘20] that is suitable for machine learning analysis. This poster discusses the application of deep learning algorithms to classify AE modes in the CO2 interferometer dataset. Preliminary results show good predictions of TAE and RSAE. Supported by the U.S. Department of Energy under DE-SC0020337, DE-FC02-04ER54698 and National Science Foundation under 1633631.