Multi-Timescale Dynamics Model Bayesian Optimization for Plasma Stabilization in Tokamaks

We present a multi-scale Bayesian optimization approach for avoiding m/n=2/1 tearing modes in high-qmin scenarios. Modelling tearing instabilities in plasma is challenging due to poorly understood physics and fluctuations in the plasma dynamics stemming from, e.g., hardware changes or impurities. To overcome these limitations, our method integrates a high-frequency data-driven dynamics model with a low-frequency Gaussian process. By updating the Gaussian process between experiments, the method rapidly adapts to new data, refining the predictions of the less reliable dynamics model. Results from an experiment session at DIII-D show a 50% success rate, which represents an improvement of 117% compared to past shots with similar configurations. Work supported by US DOE under DE-FC02-04ER54698, DE-SC0024544 and DE-SC0015480.