How to Extract Information and Knowledge from Fusion Massive Databases

The need to understand and control the dynamics of reactor grade fusion plasmas requires the analysis of increasing amounts of data, which at JET can reach easily the level of several GBytes per shot. Therefore a series of new approaches are being pursued to store the data and to retrieve the required information. They range from loss less data compression techniques, to wavelets and Structural Pattern Recognition methods. Since the information available is very often affected by high level of uncertainties and the phenomena to be studied are complex and nonlinear, the inference problems in this field of plasma physics are particularly delicate. Even in this perspective innovative solutions are under development. In particular a range of Soft Computing approaches have already been implemented at JET. The most successful are Bayesian statistics for the integration of diagnostic measurements, Data Mining techniques to study the nonlinear correlation of various variables and Fuzzy Logic to include the knowle...