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

Machine learning algorithms often struggle to control complex real-world systems. In the case of nuclear fusion, these challenges are exacerbated, as the dynamics are notoriously complex, data is poor, hardware is subject to failures, and experiments often affect dynamics beyond the experiment's duration. Existing tools like reinforcement learning, supervised learning, and Bayesian optimization address some of these challenges but fail to provide a comprehensive solution. To overcome these limitations, we present a multi-scale Bayesian optimization approach that 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 dynamical model. We validate our approach by controlling tearing instabilities in the DIII-D nuclear fusion plant. Offline testing on historical data shows that our method significantly outperforms several baselines. Results on live experiments on the DIII-D tokamak, conducted under high-performance plasma scenarios prone to instabilities, shows a 50\% success rate — marking a 117\% improvement over historical outcomes.

Controlling complex systems like nuclear fusion reactors is a major challenge for machine learning, mainly because the systems are unpredictable, data is limited, and things can go wrong during or after an experiment. Traditional machine learning methods don’t fully solve these issues.In this work, we introduce a new method that combines two types of machine learning models: one that learns from fast, real-time data and another that improves between experiments by learning from overall trends. This combination helps us make better decisions, even when conditions change or data is uncertain.We tested our method on a real fusion reactor (DIII-D) to prevent a common type of instability that disrupts experiments. First, we showed it works well on simulated experiments using past data. Then, we ran it in live experiments on a real tokamak and achieved a 50\% success rate— marking a 117\% improvement over historical outcomes.