Learning Plasma Dynamics And Robust Rampdown Trajectories With Predictfirst Experiments At Tcv Allen M Wang Alessandro Pau Cristina Rea Oswin So Charles Dawson Olivier Sauter Mark D Boyer Anna Vu Cristian Galperti Chuchu Fan Antoine Merle Yoeri Poels Cristina Venturini Federico Felici Stefano Marchioni by Allen M. Wang & Alessandro Pau & Cristina Rea & Oswin So & Charles Dawson & Olivier Sauter & Mark D. Boyer & Anna Vu & Cristian Galperti & Chuchu Fan & Antoine Merle & Yoeri Poels & Cristina Venturini & Federico Felici & Stefano Marchioni instant download after payment.

The rampdown phase of a tokamak pulse is difficult to simulate and oftenexacerbates multiple plasma instabilities. To reduce the risk of disruptingoperations, we leverage advances in Scientific Machine Learning (SciML) tocombine physics with data-driven models, developing a neural state-spacemodel (NSSM) that predicts plasma dynamics during Tokamak à ConfigurationVariable (TCV) rampdowns. The NSSM efficiently learns dynamics from amodest dataset of 311 pulses with only five pulses in a reactor-relevant highperformance regime. The NSSM is parallelized across uncertainties, and reinforcement learning (RL) is applied to design trajectories that avoid instabilitylimits. High-performance experiments at TCV show statistically significantimprovements in relevant metrics. A predict-first experiment, increasingplasma current by 20% from baseline, demonstrates the NSSM’s ability to makesmall extrapolations. The developed approach paves the way for designingtokamak controls with robustness to considerable uncertainty and demonstrates the relevance of SciML for fusion experiments.