Several kilowatt litz wire inductor has wide application, including automotive, renewable energy, household power supply, etc., which has a promising market for the next ten years. The traditional method of inductor loss calculation is a compromise between simplicity and accuracy. For example, iGSE (improved generalized Steinmetz equation) is commonly seen in core loss calculation, and the Dowell equation is commonly seen in high-frequency winding loss calculation. However, these methods, although simple and clear in the calculation, are not accurate. A neural network has been proven a powerful tool for modeling. In this paper, the authors try to model the inductor loss – core and winding loss – in a neural network approach. For core loss, Magnets dataset and three layers neural networks are used. Because magnetic bias is not included in the dataset, this model is only suitable when bias is absent. For winding loss, the sequence-to-sequence transformer network is introduced. The results show much-improved accuracy but with increased calculation complexity. So this neural network approach is suitable for higher accuracy design. Finally, the authors open-sourced the inductor loss code. Users just need to provide the necessary design parameters and working points to get inductor loss in their design.

inductor,core loss,windin,winding loss