Machine learning for tokamak scenario optimization: combining accelerating physics models and empirical models

Between-shots and real-time actuator trajectory planning will be critical to achieving high performance scenarios and reliable, disruption-free operation in present-day tokamaks, ITER, and future fusion reactors. These tools require models that are both accurate enough to facilitate useful decision making and fast enough to enable optimization algorithms to meet between-shots and real-time deadlines. While state-of-the-art integrated modeling codes come close to the accuracy and completeness needed for these applications, they are too computationally intensive. To address this problem, a novel accelerated simulation capability has been developed by applying machine learning techniques to both empirical data and physics-based simulations, enabling profile and equilibrium predictions at real-time relevant time scales. Coupled with numerical optimization schemes, the results provide a fast tool to design experiments or guide exploration of operating space. Use of AI/ML enables models to execute fast enough for use in real-time applications while maintaining high accuracy. The approach uses physics models for phenomena that are well described by models (e.g., NUBEAM), and empirical data where modeling is lacking (e.g., electron transport). An ensemble of models is generated to provide an indication of uncertainty. The optimization approach has been applied to propose optimal beam power, plasma shaping, and plasma current to achieve a target profile evolution.