Quasistatic hysteresis modeling with feed-forward neural networks: Influence of the last but one extreme values

A technique based on feed-forward neural network (FFNN) for modeling rate-independent scalar magnetic hysteresis is presented in this paper. The neural network discussed here is inspired by several papers presented in the literature. The training set is obtained by a Jiles–Atherton model, just to verify the feasibility of the method and to prevent measurements difficulties. We choose a FFNN model and in order to improve its accuracy and its ability to generalize, we make a little modification that can avoid some problems. Finally, a modification of the last model by adding the last but one extreme value as input of the FFNN is discussed.