Data-driven surrogate diagnostic for kinetic profile reconstruction on TCV

In order to efficiently operate a tokamak in a stable manner it is essential to enhance the ability to control the plasma state. In this context, kinetic profiles are paramount to inform us about the plasma state as well as for the development of advanced controllers. Nonetheless, the limited availability of diagnostics, potential acquisition failures and insufficient sampling rates pose significant challenges both for real-time control and offline analysis. Thus, there is great interest in the construction of observers based on a surrogate version of a diagnostic to complement or provide the related signal. Different formulations can be investigated, with data-driven methods being one of the possibilities. We propose such an approach for building surrogate diagnostics on TCV and demonstrate it in a first use case by reconstructing toroidal rotation and ion‑temperature profiles from charge exchange spectroscopy measurements. We focus on developing a model that robustly reproduces these profiles, while also being capable of providing a notion of uncertainty on the output. By employing such a method, we aim to identify regions of our parameter space which are lacking coverage. Subsequently, we use this information to guide new experiments, presenting an iterative procedure to maximize the information gain of each discharge when exploring this parameter space. Early studies for real-time implementation are conducted.