Recent results on structural pattern recognition for Fusion massive databases

Abstract

Physics studies in fusion devices require statistical increasing significantly. The longer the pulse, the more analyses of a large number of discharges. Given the complexity tedious is the manual pattern search. Second, the rapid of the plasma and the non-linear interactions between the increasing of imaging diagnostics should be considered. For relevant parameters, connecting a physical phenomenon with example, fast cameras may acquire images with a rate of the signal patterns that it generates can be quite demanding. Up hundreds of frames per second and, therefore, the manual to now, data retrieval has been typically accomplished by means selection of a representative image for a particular event of signal name and shot number. The search of the temporal becomes a cumbersome procedure. Third, it should be noted segment to analyze has been carried out in a manual way. t

Thermal event recognition applied to tokamak protection during plasma operation

“…If this system has been applied to only one critical TE, it can be easily extended to the recognition of others TE. To this end, we may take advantage of fusion massive databases [7] for the automatic selection and learning of TE attributes by means of data mining techniques. Another way to improve the system performance is to take into account the uncertainty on the algorithm outputs in the system decisions.…”

Overview of intelligent data retrieval methods for waveforms and images in massive fusion databases

“…For example, one cluster consists of all frames with all coefficients being 0 (black images). Other cluster contains the frames whose unique pixel with coefficient greater than 0 is pixel (2,3). Another cluster is made up of all frames with positive coefficients in pixels (i, j), i = 2, 3, 4, j = 2, 3.…”

“…The new clusters form the second layer. Although the clustering refinement could continue in order to improve the classification system, no more than two layers have been used in the application of the method to different signal collections of the TJ-II stellarator and the JET tokamak databases [3]. The searching process is accomplished in two steps.…”

Event recognition using signal spectrograms in long pulse experiments

“…Based on these classification systems, which enable the structured storage and retrieval of signals through database management systems using physical criteria in the query, parameter characterizations have been proposed, for example, morphological characteristics obtained through techniques of structural pattern recognition. 1 Various techniques of artificial intelligence, as well as other advanced treatments of signals, have been used to implement automatic classifiers. Some of these specialize in the prediction of the relevant disruptions.…”

“…1 As the duration of discharge increases, the analysis of these signals in real-time and under more complex functional requirements becomes more relevant. The acquisition and processing systems applied in this context should provide simple models.…”