Reconstruction of Tokamak Density Profiles Using Feed-forward Networks
in Aleksander, I. and Taylor, J. G. (Eds.), Artificial Neural Networks, Proceedings ICANN'92, Brighton, U.K.
1992, Vol 2 | Aleksander, I. and Taylor, J. G. (Eds.), Artificial Neural Networks, Proceedings ICANN'92, Brighton, U.K. edition
The tokamak is currently the principal experimental system for research into the magnetic confinement approach to controlled fusion. Hydrogen gas is raised to very high temperatures inside a toroidal vacuum vessel, and the resulting plasma is confined by a complex system of magnetic fields. Measurements of the electron density inside a tokamak can be made using laser interferometry, which gives line-integral information along chords through the plasma. Extraction of spatially local information from this line integral data represents an ill-posed inverse problem. In this paper we describe a novel approach to the solution of this problem, based on feedforward networks, and we show that it leads to improved accuracy of reconstruction compared with conventional techniques. A software implementation of the trained network has been installed at JET and will be used on a routine basis for profile reconstruction.