Machine learning surrogate model for rapid prediction of electron cyclotron heating and current drive in tokamaks

The challenges involved in iterative optimization of plasma scenarios and engineering specifications for next-generation fusion devices necessitate the use of reduced models to accelerate the design process. High-fidelity ray-tracing codes, such as TORAY, offer considerable advantages in runtime for obtaining electron cyclotron (EC) heating and current drive (H/CD) profiles, yet large parameter scans for H/CD optimization are still time-consuming. We utilize the TokDesigner workflow to train a machine learning-based surrogate model for EC H/CD radial profiles based on integrated modeling using TORAY coupled to the Integrated Plasma Simulator (IPS)–FASTRAN framework. The surrogate model utilizes Gaussian process regression (GPR) to predict key EC H/CD profile characteristics with uncertainty quantification (UQ) for cases based on plasma and EC launch parameters for a Compact Advanced Tokamak (CAT) design point. This surrogate model expands upon previous work [Irvin, A. M. et al. Fusion Science and Technology, 1–15. (2025)] through the application of machine learning to the established method in order to obtain more accurate predictions in orders of magnitude less runtime. Further application to an experimental dataset is underway for verification and optimization of this surrogate modeling method.