A Comparative Analysis on Different Deep Neural Network Models for Magnetic Hysteresis with Distorted Excitation Waveforms

Hereby presented is a comparative analysis of different deep neural network models aimed at simulating magnetic hysteresis under distorted excitation waveforms. Specifically feed forward neural networks (FFNN), recurrent neural networks (RNN), and nonlinear autoregressive exogenous (NARX) networks have been chosen to be evaluated due to their distinct capabilities in capturing non-linear relationships and temporal dependencies, making them suitable for the complex and dynamic nature of hysteresis phenomena. Utilizing the Preisach model to generate training and validation data, the study highlights the ability of neural networks to model complex magnetic behaviors. The final results indicate that FFNNs achieve the highest accuracy with an average error of 3%, outperforming RNNs and NARX networks. The findings underscore the suitability of FFNNs for applications where training datasets are limited or computational resources are constrained.