Abstract: In order to solve the problem of low efficiency of two-dimensional fluid-thermal coupling simulation of oil-immersed transformer, a parallel calculation method based on hybrid finite element method was proposed. Firstly, the serial calculation method of dimensionless least squares finite element method and upwind finite element method is implemented by C++ language in Visual Studio 2019. Then, the parallel computation of fluid field is realized based on graphic processing unit (GPU), and the parallel computation efficiency of different GPU cards under different grid conditions is compared and analyzed according to the single differentiated turns model. The analysis results show that the larger the data scale is, the more the GPU card stream processors will be, and the better the parallel effect will be. Secondly, based on Intel MKL and open multi-processing (OpenMP), the parallel computation of two-dimensional temperature field is realized, and the influence of different number of grids on the parallel efficiency is compared and analyzed. Finally, a hybrid parallel calculation method based on different simulation conditions is proposed and applied to the analysis of the two-dimensional temperature rise hot spot of the winding model of large oil-immersed transformer. The results show that, compared with the serial program, the speedup ratio of the hybrid finite element parallel calculation method reaches 69.5. The experimental results further verify the accuracy of the parallel calculation results, and the research results lay a foundation for the rapid calculation of the flow-thermal coupling problem of large oil-immersed transformers.