Neural network for representation of hysteresis loops

Abstract

Presents a mapping model for the representation of symmetrical B/H characteristics over the whole of the B/H plane, based on neural networks taught by backpropagation. The model could not be achieved accurately by just using a symmetrical saturated hysteresis loop to simulate a smaller hysteresis loop. Eleven experimentally obtained hysteresis loops from over the whole of the B/H plane were used to train neural networks. These are good enough to represent all nonlinear hysteresis characteristics to meet the needs of an engineering calculation, e.g. a transient performance analysis of a current transformer or voltage transformer. The simulation accuracy depends ultimately on the accuracy of the experimental data. The computed results using this model have shown a good agreement with measured data.