The use of neural network approximation models to speed up the optimisation process in electrical impedance tomography (Q839172): Difference between revisions

The considered mathematical model for the cavity detection problem consist of the Laplace equation on E - D and three boundary conditions on the boundary of E and D, where E and D are two domains with Lipschitz boundary and D is completely included in E. The case of circular cavities will be investigated only. If a solution for the Laplace equation and two boundary conditions can be computed, the third boundary condition will be fulfilled if a certain quadratic functional gets minimized. This way, the mathematical model for the cavity detection is transformed into an optimization problem. For the optimization process one uses ES , the evolutionary strategy. Further, the purpose is to use neural networks technology for approximating the objective function. MLP (MultiLayer Perceptrons) as well as RBF (Radial Basis Functions) networks are used for constructing the reduced model. The MLP consists of one hidden layer, N input and one output signals, a sigmoid activation function and is trained by backpropagation. The RBF network has Gaussian basis functions, one output and the centroids are clustered by Kohonnen training. Some numerical examples are presented to prove the way ES is accelerated by using MLP or RBF approximating models.