Neural Network-based Medium Frequency Transformer Electric Field Modeling and Sensitivity Analysis

The insulation of medium frequency transformers has become an important bottleneck restricting the further improvement of its reliability and power density. However, the insulation electric field intensity is closely related to the shape and geometric parameters of the magnetic core and windings, and it is difficult to perform accurate physical analytical modeling based on Maxwell’s equations. In this paper, based on the neural network data-driven algorithm, in view of the actual geometry of medium frequency transformer, comprehensively consider the influence of the full design variables on the local maximum electric field strength, a general insulating electric field analytical model is established. The model considered the calculation accuracy and calculation efficiency, and the accuracy can be up to 97.8%, and the single calculation time is 0.86s. Finally, based on the Sobol algorithm, the analysis shows that in addition to the insulation distance, the diameter of the low-voltage winding has a higher sensitivity to the local maximum electric field strength.