Bayesian approach to autonomous kinetic profile analysis in KSTAR: Development and initial results

The unique superconducting magnet design of KSTAR presents challenges for diagnostics and heating near the vessel, complicating data interpretation and requiring substantial resources. To address these limitations, we have developed autonomous AI-based fitting techniques for kinetic profile analysis. Our approach leverages recent advancements in AI/ML, employing a combination of Bayesian methods, Gaussian processes, and Markov Chain Monte Carlo (MCMC) for both profile fitting and outlier detection. By cross-checking multiple diagnostics and using synthetic diagnostics, we infer the most probable kinetic profile fitting. Initial results demonstrate improved fitting accuracy and significantly reduced processing time. The system is scheduled for distribution during the 2024 KSTAR campaign, with plans to implement between-shot fitting result archiving. This innovative approach aims to enhance the efficiency and reliability of kinetic profile analysis in KSTAR, potentially offering insights for other fusion devices facing similar diagnostic challenges.