Data-driven learning for disruption prevention and performance optimization
Cristina Rea
Massachusetts Institute of Technology
Plasma Science and Fusion Center
This seminar will focus on the development of interpretable Machine Learning (ML) driven solutions for two critical challenges in magnetic confinement fusion: (1) real-time monitoring of proximity to plasma stability boundaries and (2) the optimization of plasma trajectories. Modeling and experimental results carried out at DIII-D and TCV will be presented. At the core of these advancements is a shift towards data-efficient Scientific Machine Learning: this hybrid modeling approach enables interpretability by coupling the known physics models to ML corrections to advance the reliability, efficiency, and scalability of ML-enabled solutions for both existing and next-generation fusion devices.