Searching for patterns in TJ-II time evolution signals

Since fusion plasma experiments generate hundreds of signals, it is important for their analysis to have automatic mechanisms for searching for similarities and retrieving specific data from the signal database. This paper describes a technique for searching in the TJ-II database that combines support vector machines and similarity query methods. Firstly, plasma signals are pre-processed by wavelet transform or discrete Fourier transform to reduce the dimensionality of the problem and to extract their main features. Secondly, support vector machines are used to classify a set of signals by reference to an input signal. Finally, similarity query methods (Euclidean distance and bounding envelope) are used to search the set of signals that best matches the input signal.