Search and retrieval method of similar plasma waveforms

Abstract

Abstract Fusion plasma experiments often produce many channels of time-varying waveforms, and the assistance of a computational method to find similar waveforms could accelerate data analysis. To accelerate computer aided search, a so-called “thumb-indexed” reference will be required in the waveform database. Fourier series coefficients have been chosen for “R-tree” multi-dimensional indexing. As one waveform will be translated into one point in the multi-dimensional space of the Fourier series, the similarity between two points can be calculated as their Euclid distance in this space. For slowly varying waveforms like large helical device (LHD) bolometer signals, a two-step search turned out to provide the best calculation speed of the relative distances, the first step uses the primary (0, ±1, ±2, …, ±k)=2k+1 Fourier complex coefficients, and the second uses the other 2m ones whose amplitudes are the largest. Evaluation tests varying k and m have given the following best pair of values 2k+1=5 and 2m=4 for retrieving waveforms similar to those of the bolometer. The speed and precision of the method were also confirmed. This method makes the possibility to directly search similar phenomena in huge waveform databases realistic.