Analysis of some numerical models of hysteresis loops

Five models including the neural network method proposed by the authors for the representation of the hysteresis phenomena in magnetic materials have been examined and compared with reference to their use in the design of current and potential transformers. Most of the models referred to in this paper are of the linearized type and so cannot avoid introducing errors. The neural network approach handles the nonlinearities well and the accuracy is good, but it requires extensive test data and a computer intensive learning period. In use, the neural network approach is fast and gives an accuracy of better than 1% over the whole B/H plane