Optimal identification using feed-forward neural networks

Abstract

In this work we present new approaches for the optimal identification of nonlinear systems. We optimize different parameters of feedforward neural networks and of the learning schedule backpropagation by the use of global search methods like genetic algorithms and simulated annealing. We achieve a global increment of their learning capability thereby enlarging the generalization capability and reducing the amount of learning speed. The result is a more reliable and robust model for nonlinear systems.