May 28, 2026

Paper on Machine-Learning-Based Plasma Confinement State Classification Published in Nuclear Fusion

Our paper, “Plasma confinement state classification in fusion power plants: Profile reflectometer and ensemble diagnostics,” has been published in Nuclear Fusion. The work was carried out by Randall Clark (UC San Diego, Orlov Lab), Vacslav Glukhov, Georgy Subbotin, Maxim Nurgaliev, Aleksandr Kachkin (Next Step Fusion, S.a.r.l.), Lei Zeng (UCLA), and Dmitri M. Orlov (UC San Diego).

The study investigates machine-learning-based plasma confinement state classification using a limited set of diagnostics relevant for future fusion pilot plants (FPPs). In reactor environments, diagnostic availability is expected to be strongly constrained by survivability, maintainability, and limited port access. To address this challenge, the authors developed a confinement mode classifier using DIII-D profile reflectometer (PR) data and combined it with a previously developed electron cyclotron emission (ECE)-based classifier into an ensemble framework. The PR-based model achieved approximately 97% classification accuracy between L-mode and H-mode plasmas, while the ensemble model achieved greater than 99% accuracy on randomized tests and approximately 96% accuracy on chronologically separated sliding-window tests designed to evaluate robustness against previously unseen plasma conditions. The work demonstrates that robust plasma state identification may be achievable using a relatively compact set of reactor-compatible microwave diagnostics, helping support future development of plasma monitoring and control systems for fusion power plants.

The publication is available online at: https://doi.org/10.1088/1741-4326/ae72a8