Dynamic Clustering and Modeling Approaches for Fusion Plasma Signals

This paper presents a novel clustering technique that has been applied to plasma signals to show its utility. It is a general method based on a partitioning scheme that has been proven to be efficient for purposes of analysis and processing of fusion plasma waveforms. Moreover, this paper shows how the information given by the clustering can be used to produce a concise and representative model of each class of signals by applying different modeling approaches. Neuro-fuzzy identification and time-domain techniques have been used. These models allow the application of procedures to detect anomalous behaviors or interesting events within a continuous data flow that could automatically trigger the execution of some experimental procedures. Previously, an in-depth analysis and a preprocessing phase of the waveforms have been carried out. These procedures have been applied to plasma signals of the TJ-II Stellarator fusion device with encouraging results.