Bayesian Integrated Data Analysis of ITER Magnetic Equilibrium and Kinetic Profiles

We have created a flexible Integrated Data Analysis (IDA) framework that combines multiple ITER diagnostics. Forward models for ITER’s magnetics, Thompson scattering, TIP, DIP, CXRS, and XRCS diagnostics were developed. Neural networks were used to replace computationally expensive equilibrium calculations. Low dimensional parametrizations of the plasma profiles and magnetic flux significantly reduce complexity and allows for faster and more robust inference. With these advancements we are able to simultaneously calculate the magnetic equilibrium and kinetic plasma profiles and associated uncertainties on the order of tens of minutes.

The IDA framework was developed because ITER discharges are expected to be challenging to determine a magnetic equilibrium due to each diagnostic only providing a partial view of the plasma and their respective forward models being unable to make the usual simplifying assumptions due to the plasma regime. We, therefore, took an integrated approach to inference where each diagnostic model is simultaneously evaluated and optimized. Furthermore, we used Bayesian methods that provide a natural way of combining diagnostics and other prior information while also providing a unified result, rigorous uncertainty quantification, and a method of identifying calibration errors. The uncertainties of the inferred parameters can then be properly propagated in the calculation of derived processes such as stability analysis and other control relevant studies.