Real-Time Kinetic Equilibrium Reconstruction in Tokamak Plasmas Using Machine Learning

The development of a fast and reliable kinetic equilibrium reconstruction tool is crucial for real-time prediction and control in tokamak plasmas. In this study, we propose a machine learning (ML)-based approach, RTCAKENN, to rapidly generate kinetic equilibria using various plasma diagnostic signals. The ML models are trained with tens of thousands of time slices reconstructed through the Consistent Automatic Kinetic Equilibria (CAKE) workflow. The input signals, which include scalar independent and interdependent variables, one-dimensional profiles, and two-dimensional plasma boundary coordinates, are preprocessed and fed into a deep neural network (DNN) architecture. The DNN model encodes the input signals, extracts comprehensive latent features, and generates the desired output profiles that satisfy the kinetic magnetohydrodynamic (MHD) equilibrium. The proposed approach has the potential to enhance plasma profile-based prediction and control in tokamak experiments.