Description
Machine Learning Based Tungsten Spectroscopy in WEST Tokamak
Sadruddin Benkadda, PIIM Laboratory, Aix Marseille University–CNRS
Abstract
A machine-learning approach to tungsten spectroscopy analysis is developed using measurements from WEST tokamak plasmas. A Random Forest model is trained to predict the maximum electron temperature, denoted as Te max, from tungsten brightness spectra in the wavelength range 45–65 angstroms, measured along a mobile line of sight. The model achieves prediction errors typically below 5 percent over a broad electron temperature range from 0.5 to 4 kilo–electronvolts. Feature-importance analysis identifies both physically meaningful wavelengths and others that, although lacking direct physical interpretability, contribute to improved prediction stability and accuracy. In addition, principal component analysis is performed to investigate the relationship between spectral variance and Te max, revealing that previously unaccounted-for parameters influence the spectral shape. When combined with the wavelength-importance analysis, these results may help inform atomic structure and collisional–radiative models, leading to an improved understanding of tungsten spectral emission.
