Forecasting disruptions Instabilities by artificial neural networks / Previsão das Instabilidades de Disruptura através de Redes Neurais Artificiais
AUTOR(ES)
Kenya Andrésia de Oliveira
FONTE
IBICT - Instituto Brasileiro de Informação em Ciência e Tecnologia
DATA DE PUBLICAÇÃO
25/02/2000
RESUMO
Two-layer feedforward neural network has been used in this work to forecast the disruptive instabilities that occur in the TEXT tokamak plasma discharges. For this task, soft X-ray experimental signals were used with very promising results. It was verified that a neural net with an architecture of the type m:2m:m:1, where m is the embedding dimension of the atractor of dynamical system in focus, is usually a good initial guess in the searching process of finding the ideal architecture. A neural network with architecture 15:30:15:1 was capable, for example, to forecast the disruptive instabilities up to 4 ms in advance. This period of time is four time larger than the one obtained when magnetic signals from Mirnov coils were used. This forecasting time is quite significative and opens up the possibility of using defensive mechanisms, such as the injection of neutral particles (or pellets), the application of external magnetic fields, etc, with the objective of avoiding the occurrence of the disruptions or, at least, to minimize their harmful effects. This achievement certainly would be an important contribution to the development of the next generation fusion devices. Finally, the soft X-ray diagnostic system for the TCABR was projected and it is already being installed. This system will provide experimental signals that will be analyzed by neural networks and will be also used to identify, through tomografic image reconstructions, the regions of the plasma that have the same soft X-ray emissivity. The tomography analysis of the plasma, that will be carried out by using the signals of two soft X-ray detectors arrays, will be also very usefull for investigating the triggering mechanism of disruptions and will also allow the determination of the plasma electron temperature through the two foil absorbing method.
ASSUNTO(S)
disruptions instabilities forecasting instabilidade de disruptura neural networks redes neurais artificiais tokamak tokamak
