| Hi-index | 0.01 |
In this paper we present results from the first use of neuralnetworks for real-time control of the high-temperature plasma in atokamak fusion experiment. The tokamak is currently the principalexperimental device for research into the magnetic confinementapproach to controlled fusion. In an effort to improve the energyconfinement properties of the high-temperature plasma insidetokamaks, recent experiments have focused on the use of noncircularcross-sectional plasma shapes. However, the accurate generation ofsuch plasmas represents a demanding problem involving simultaneouscontrol of several parameters on a time scale as short as a fewtens of microseconds. Application of neural networks to thisproblem requires fast hardware, for which we have developed a fullyparallel custom implementation of a multilayer perceptron, based ona hybrid of digital and analogue techniques.