Plasma Parameter Optimization Using a Neural Network

Abstract

Our understanding of plasma transport in magnetic confinement devices is sketchy. Neither analytic nor computational models are sufficiently well developed to describe accurately the behavior of real experimental systems. Artificial neural networks (Bishop, 1994), on the other hand, are capable of modeling complex input-output relationships learned on the basis of empirical examples alone. As an initial effort to apply neural network technology to plasma optimization and modeling we have employed data taken from a small research stellarator (Jones, 1985). In the experiment which we wish to model, it is possible to vary the magnetic field used to confine the plasma, B; the neutral