Measuring the electron temperature and identifying plasma detachment using machine learning and spectroscopy

A machine learning approach has been implemented to measure the electron temperature directly from emission spectra of a tokamak plasma. This utilized neural network (NN) trained on dataset 1865 time slices operation DIII-D using extreme ultraviolet/vacuum ultraviolet spectroscopy matched with high-accuracy divertor Thomson scattering measurements temperature, Te. NN is shown be particularly good at predicting Te low temperatures (Te < 10 eV) where demonstrated mean average error less than 1 eV. Trained detect plasma detachment in divertor, classifier was able correctly identify detached states 5 99% accuracy (an F1 score 0.96) an acquisition rate 10× faster measurement. The performance model understood by examining set 4800 theoretical generated collisional radiative modeling that also used predict low-cost spectrometer viewing nitrogen visible wavelengths. These results provide proof-of-principle spectrometers leveraged can boost more expensive diagnostics fusion devices and independently as fast accurate measurement classifier.