Speaker
Description
Integrated data analysis is essential for the full exploitation of diagnostic measurements. In the past various approaches were investigated but the main techniques utilised were based on Bayesian statistics. Physics-Informed Neural Networks (PINNs) are an alternative worth addressing. Indeed PINNs constitute a new branch of artificial intelligence that gives the possibility of integrating data-driven methodology and physics equations in a very efficient way. They offer several benefits over traditional methods, such as their capability of handling incomplete physics equations, of coping with noisy data, and of operating mesh-independently. The subject of the present work consists of assessing the potential a Physics-Informed Neural Network (PINN) algorithm for reconstructing the plasma equilibrium using a multi-diagnostic approach, which includes magnetics, kinetic pressure, and interferometer-polarimeter data. This constitutes a quite severe benchmark for any strategy of integrated data analysis. Indeed in a tokamak the equilibrium reconstruction is a severely ill-posed problem. To achieve reasonably accurate results, it is therefore essential to constrain the algorithms with multiple diagnostic data. Among these, the interferometer-polarimetric measurements are some of the most valuable, as this diagnostic is one of the few that can provide information about the internal fields, even if in a line-integrated form. However, the propagation of an electromagnetic wave in a magnetised plasma requires the quantification of significant non-linear effects, which render the integration of this information into the reconstruction process anything but straightforward. Unfortunately, the linearisations and approximations implemented in the past limited the quality of the reconstructions, particularly at high fields and currents. On the contrary, the developed PINN algorithm implements a complete hot plasma model that accounts for these nonlinearities and also thermal effects, both relativistic and non-relativistic. A systematic series of tests with synthetic data demonstrates that the hot plasma model provides results consistently more accurate than those obtained with the cold-plasma approximation or linearization of the polarimetric measurements. The models derived with the PINNs have been also tested with JET data collected high current campaigns, confirming the quality of the obtained reconstructions in all the investigated experimental conditions.
