Real-time ion temperature profiles in the JET nuclear fusion experiment

A real-time system for the calculation of ion temperature profiles in the JET nuclear fusion tokamak has been developed. The process of reconstructing ion temperature profiles from measured data involves the analysis of rather complicated spectra in the visible region, emitted from several regions in the hot fusion plasma. Current analysis is based on a maximum likelihood approach, where a high-dimensional non-linear minimisation is made for each spectrum. The result of the minimisation is strongly dependent on the choice of the manually supplied initial guesses, due to near-ambiguities caused by overlapping spectral lines. Here, we use a system of MLP neural networks for the unfolding process, thus taking away the need for manually supplied initial guesses, and reaching far higher execution speeds. This together makes realtime analysis possible. Since the demands on robustness and accuracy are high, an effort has been made to optimise the precision, and several confidence measures are used, all compliant with the real-time demands on fast execution speed. The result is the first system available for real-time ion temperature profile reconstruction in a nuclear fusion tokamak.