Optimisation of intensity refractometry design and density reconstruction procedure

Abstract Intensity refractometry is a recently developed microwave technique for measuring density in plasmas exhibiting strong radial density gradients, such as those found in magnetised fusion plasmas. This technique is attractive due to its measurement accuracy, simplicity, low cost, and robustness in the fusion plasma environment. While some examples of diagnostic design and data analysis approaches have been previously presented, no systematic design study for intensity refractometry has yet been carried out. This paper presents a new, versatile numerical framework aimed at optimising the diagnostic design parameters to meet typical experimental requirements. These requirements include the location and span of the measurement region, density range, density profile shape, measurement precision, available space for in-vessel components, and the number of measurement channels. Concurrently with diagnostic design optimisation, the developed framework also optimises the density reconstruction procedure, which is crucial for diagnostic performance. The machine learning optimisation framework employs Bayesian optimisation, specifically using the Tree-Structured Parzen Estimator algorithm within Optuna, an open-source hyperparameter optimisation framework. A challenging, high-dimensional, constrained, and batched optimisation problem was formulated—an area scarcely addressed in existing literature. This presented multiple challenges, including issues with convergence, and necessitated the exploration of various approaches suitable for such complex optimisation problems.