Adaptive models and methods

Development of modern mathematical methods and convenient software tools for processing diagnostics data is an important direction of research in controlled fusion. Such methods and tools help the automation of huge data analyses and the prompt extraction of valuable information about plasma behavior.

One approach to the solution of many complicated problems in plasma physics and controlled fusion involves modern adaptive data mining tools. The efficiency and productivity of this kind of tools are based on their ability to extract very valuable information from seemingly chaotic data without deep knowledge of underlying physical laws, which in fact can be too complicated or still unformulated. Indeed, the extracted information can help the formulation of these laws and the making of correct decisions.

Recently the following relatively new to fusion research techniques for plasma diagnostic data processing that were developed: plasma boundary reconstruction using video images, optimization of real or numerical experiments with artificial neural networks, application of support vector machine to the classification of plasma pulses, processing of magnetic diagnostics data using hidden Markov models, clustering and navigation in the database of graphical information with Kohonen self-organizing maps, reconstruction of the distribution of light source using high-resolution plasma images.

Examples of successful solution of different problems in fusion prove the high efficiency of the methods and motivate further applications.

Details of adaptive approaches can be found, for example, in “A.A. Lukyanitsa, F.S. Zaitsev, A.G. Shishkin, S.V. Nosov, A.V. Morozov, F.M. Zhdanov, V.V. Zlobin. Data mining methods in controlled thermonuclear fusion. The First Korean-Russian Workshop on Data Mining. - Moscow: MAX Press, 2007. P. 17-25.”