Real time analysis of multichannel data in tokamaks

Four different techniques for the fast analysis of multichannel data in plasma physics are discussed. All four of these techniques are general and sufficiently fast to be used in real time applications. Function parametrization, canonical correlation analysis and a neural network of the multilayer perceptron (MLP) type are compared with a unique linear mapping based on a singular value decomposition, which is used as a reference. Applications deal with the identification of the plasma boundary and some global plasma parameters in the DIII-D and the Tore Supra tokamaks by using magnetic measurements. The results of an MLP-1 neural network, employed for the real time plasma position determination in Tore Supra, are presented