Neural network approach to the solution of the inverse problem of plasma boundary determination (Q1269953)

This work presents the formulations of direct and inverse equilibrium problems, a numerical method for solving equilibrium problems with a free boundary according to Lackner's procedure, and the characteristics of the system of confining currents on modern experimental devices. The main emphasis is on the possibility of using the neural network approach in the solution of inverse equilibrium problem; in particular, we give a detailed description of obtaining the neural network training, of the neural network itself, and the numerical investigation of the solution of the reconstruction problem by means of neural networks. Additionally, we give a new regularization method for training of multilayer-type neural networks.