A neural network approach to evaluate density profiles from reflectometry in ASDEX Upgrade discharges with internal transport barriers

Abstract

In next step devices it is expected that reflectometry can be used as an alternative to magnetic systems in the control of plasma position and shape. This is particularly important in long discharges when the accumulated errors of magnetic signals may be quite significant. This is beyond the present application of reflectometry and puts new requirements on the diagnostic, namely automatic analysis of reflectometry data, real-time data processing, and high reliability. A key step is to demonstrate the potentialities of real-time analysis in present reflectometry systems. With that purpose, we propose a neural network approach to process simulated and experimental data measured with reflectometry on the ASDEX Upgrade tokamak. The study shows that the neural network approach has the potential to meet the tight timing requirements of control applications with sufficient accuracy, provided that realistic profiles are used in the training. First tests using ASDEX Upgrade reflectometry data are promising.