Abstract: High-frequency transformer is the core component of power electronic transformer to realize electrical isolation and voltage conversion of new energy AC/DC system, which has the characteristics of small volume and high power density. Affected by high-frequency skin effect and proximity effect, the winding loss of high-frequency transformer is generally large. In addition, due to its compact structure and insufficient heat dissipation capacity, it is easy to cause the local temperature rise of the equipment to exceed the limit. It is of great significance to calculate winding loss accurately and efficiently for the optimal design and reliable operation of high-frequency transformers. This paper analyzes the microscopic honeycomb structure of the Litz-wire, extracts the basic periodic unit to simplify the electromagnetic field analysis, and proposes the winding electromagnetic field homogenization simulation model and loss calculation method based on the equivalent medium theory. The results show that the homogenization simulation model has high accuracy and computational efficiency. By taking the simulation results of the refined two-dimensional model as a reference, the simulation error of the magnetic flux density and the electric field strength is less than 0.5%, the computational efficiency is improved by about 12 000 times, and the running memory requirement is reduced by 30.28 times. The accuracy of the proposed loss calculation formula is similar to that of the mainstream model in the middle- and low-frequency bands, while the error in the high-frequency band is 20% lower than that of the Tourkhani model. The homogenization model proposed in this paper significantly reduces the modeling difficulty and calculation amount of the electromagnetic field simulation of the Litz-wire winding, solves the problems of high-frequency eddy current simulation difficulty and low loss calculation accuracy, and provides theoretical and model supports for the structural optimization design of the high-frequency transformer.