Data-Driven Prediction and Control of Alfvén Eigenmodes at DIII-D

We have previously developed a data-driven model based on reservoir computing networks (RCNs) to detect, classify, and localize Alfvén eigenmodes in a database of over 1000 DIII-D discharges. The model looks at 40 Electron Cyclotron Emission (ECE) signals, sampled at 500kHz, to distinguish AE activities such as Low-frequency modes (LFMs), Beta-induced Alfvén eigenmodes (BAE), Reversed-Shear Alfvén eigenmodes (RSAE), and Toroidal Alfvén eigenmodes (TAE).

Here, we report on follow-up research to develop multimodal machine learning models to predict and control the AE modes based on the current plasma profiles, diagnostics, and actuators. The model is trained on various input data such as ECE, neutron rate, electron density, plasma pressure, injected neutral beam power and ECH. Our preliminary results show the potenital of the proposed model in predicting the probability of pronounced AE activities 200ms in advance. To control the AE activities in real-time, such a model can be used to tune the actuator values by predicting the evolution of AE modes based on the current plasma conditions and different proposals for actuator values.

The models will be deployed for integration in the DIII-D plasma control system (PCS) to be tested in the 2022-23 campaign.