AI for real-time fusion plasma behavior prediction and manipulation

Princeton's Plasma Control Group is utilizing machine learning to enhance real-time control of plasma behavior in tokamaks. Their work includes new methods for managing Edge Localized Modes, tokamak profile control, monitoring plasma behavior with high-resolution diagnostics, and reducing or up sampling diagnostic data using data-driven approaches. Notable achievements include a machine-learning model predicting plasma states in under 100 microseconds, and ML applications enabling better understanding and operation of tokamaks and future fusion reactors.

• Tokamaks utilize several actuator types for plasma control.
• Super-resolution data captures detailed plasma evolution.
• ML models can detect Alfven-Eigen modes in the plasma.
• Neural networks recreate diagnostic signals in simulations.
• Real-time models facilitate stable divertor radiation detachment.