Real-time extraction of plasma equilibrium parameters in KSTAR tokamak using statistical methods

Abstract

To improve inherent shortcomings of statistical methods and apply them to the extraction of plasma equilibrium parameters in a fast timescale for real-time plasma control, new concepts of statistical methods such as principal component analysis-based neural network (NN), functional parametrization (FP)-based NN and double network are introduced by modifying NN and FP. These new methods are benchmarked and compared with the conventional techniques of NN and FP in a simple single-filament system. As a result of their applications to identification of plasma equilibrium parameters in the Korea Superconducting Tokamak Advanced Research tokamak, particularly, the double network concept among them has successfully achieved the improvement of drawbacks in the conventional methods. It is shown that more reliable results from the double network method can be obtained by combining several different statistical treatments as a primary network. Even in the case of nonoptimized methods united as a primary network, quite acceptable results can be achieved in the double network method. © 2001 American Institute of Physics.