Artificial Neural Networks for Data Analysis of Magnetic Measurements on East

The problem of the reconstruction of the parameters characterizing the plasma shape in a tokamak device is of paramount importance both for present day experiments and for future reactor. The plasma shape can only be evaluated by diagnostic data, such as poloidal flux and magnetic field measured respectively by the flux loops and magnetic probes located on the vacuum vessel outside the plasma. The aim of the present paper is to take a step forward in the application of the neural network approach for the identification of non-circular plasma equilibrium and data analysis for the problem of the optimal location of a limited number of magnetic sensors. We have adopted a machine learning method, back-propagation neural network, to analyze the magnetic diagnostic data. The database has been generated by means of a specially adapted version of an MHD equilibrium code EFIT with reference to the EAST geometry and stored in the EAST mdsplus database. The network uses external magnetic measurements as input data and the selected plasma parameters as output data to train and test. Then a novel strategy is implemented for the selection of the optimum location of a limited number of magnetic probes based data analysis of the network. The average accuracy of the identification procedure is quite good (e.g., the maximum relative error is 0.260 % of internal inductance), with a contrast of the computation results of EFIT as desired output. It has been shown that the degradation of the performance is rather small (e.g., RMS error of minor radius vary from 4.307 to 4.765 %) when the number of magnetic probes is reduced by nearly half.