Machine learning models for inverse and lateral problems of lower hybrid current drive

This presentation describes the development of machine learning models to solve the inverse and lateral problems for lower hybrid current drive (LHCD) simulations. The inverse problem in physics involves solving for the set of causal factors that result in a set of observations. Recent work [Wallace et al, JPP 2022] has shown that machine learning surrogate models can accurately reproduce the current density and power absorption profiles calculated by GENRAY/CQL3D (a suite of ray-tracing/Fokker Planck solvers) given a set of 0D input parameters describing the plasma shape, temperature, density, and LH waves based on a database of simulations. In the inverse problem context, the observations are current density and power absorption profiles calculated by GENRAY/CQL3D, and the causal factors are the input parameters to the codes. An analogous problem, which we call the lateral problem, relates one output of GENRAY/CQL3D (e.g. the hard X-ray emission synthetic diagnostic) to another output (e.g. the current density profile). This presentation focuses on the application of ML-based models to the inverse and lateral problems in situations where analytic or iterative solutions are impractical.