Application of genetic algorithms in the analysis of active databases (Q1285465)

As a consequence of instability in the plasma behavior there is a quick loss of the magnetic and thermic power of charges in tokamak systems. It is important therefore to develop effective methods which allow to recognize and analize unstable regions in many-dimensional cases. Two methods of pattern classification based on genetic algorithms are proposed in the article. The first one using the projection method permits to select a group of the plasma parameters carrying a base information which is necessary for diagnostics of the plasma behavior. The second method is based on construction of dividing hyperplanes which are approximating nonlinear bounds between stable and unstable regions of the many-dimensional space of the plasma parameters. A real database with the values of seven global plasma parameters tested in 320 experiments on the COMPASS-D tokamak system (Culham laboratory, UK) is analized in this article.