Reduction of magnetic confinement fusion data for data mining applications

We develop a practical, structured analysis of multi-channel time series measurements where the main interest lies in the coherent temporal fluctuations and spatial structures and their time dependence. The current approach to most large scale plasma experiments, toka-mak and stellarators alike, is the quest for the experimental data taken under optimal conditions for each study. These data are then anal-ysed in detail and sometimes distributed in a reference database such as the toka-mak profile database of the filter 1D Modelling Working Group. While these results are important for our understanding of future fusion devices, they do not provide easy means to support the evidence based on statistical ensembles. The raw data which are not accessible to simple search queries, are usually kept in large data repositories. At H-1, we routinely log the global experimental parameters in a summary database which is stored in a easily accessible database. In order to facilitate statistical analysis and the search for a wide class of magnetic phenomena, we developed a data processing procedure that reduces the raw signal of an array of Mirnov coils at the H-1 into a series of feature descriptors in time-frequency space which are stored in an SQL-accessible database, which can be used together with the summary database.