Machine Learning Approaches to Plasma State Mode Classification via Reactor Relevant Diagnostics at DIII-D

We have demonstrated successful L-H mode classification using supervised learning models trained exclusively on reactor-relevant diagnostics from DIII-D. These include the Electron Cyclotron Emission, Visible Filter scopes, and the Radial Interferometer-Polarimeter (RIP), representing a reduced diagnostic set expected to remain viable in fusion pilot plants. Classification of the confinement regime in tokamak, L or H mode, is a critically important task for the future reactor/FPP, necessary to maximize fusion power and maintain safety and stability of the plasma. We achieved a reasonable success by passing the temperature from ECE through a feature extractor and then utilizing Gradient Boosting Classifiers and have plans to extend this to include density from the Profile Reflectometer. Visible Filter scope and RIP spectrograms can reveal signatures of ELMs and LH transitions that can be used to enhance classifier performance. These results indicate that robust plasma state classification is achievable even under severe diagnostic constraints. We present a survey of these approaches with these diagnostics to identify the most promising and accurate tools. Future work will integrate these models into real-time control frameworks and extend them to other reactor scenarios.