Machine Learning Prediction of the High-Field Side Scrape-Off Layer Density and Optimization of DIII-D HFS LHCD Antenna Loading

High-field side lower hybrid current drive (HFS LHCD) is a potential candidate to provide efficient, non-inductive, off-axis current drive in tokamaks for steady-state operation, stability control, and/or access to advanced tokamak scenarios. LHCD is sensitive to scrape-off layer (SOL) conditions local to the launcher. A high-field side reflectometer has been installed on DIII-D to characterize the HFS SOL and provide accurate measurements of LHCD-SOL interaction. The reflectometer operates in the 6-19 GHz range in O-mode polarization, corresponding to a measured density of 4.5×10¹⁷ – 4.5×10¹⁸ m^-3, with future extension possible to 6-27 GHz. Using a database of HFS SOL measurements from this reflectometer, machine learning techniques (NN, XGBoost) have been used to predict HFS SOL density profiles from plasma shaping parameters and global plasma parameters such as plasma current and density. This enables accurate prediction of HFS LHCD antenna loading, which is calculated using 3-D finite element full-wave codes, from controllable shot parameters. Bayesian Optimization is then used to search the parameter space and optimize the antenna loading for maximum current drive efficiency.