Abstract: Understanding the temperature situation of windings of oil immersed power transformers is the key to ensuring their operational stability, and it is also an inevitable requirement for digital twin analysis of oil immersed transformers. In order to obtain the steady-state temperature of transformer windings more quickly, this paper proposes a reduced-order calculation model based on proper orthogonal decomposition (POD) and radial basis function response surface method including linear polynomial (RBFLP). First, the article discusses the order reduction characteristics of the POD method and designs an adaptive snapshot matrix acquisition method based on the leave one out (LOO) method cross validation to improve computational accuracy and efficiency. Then, the response surface methodology is used to establish the correlation between POD modal coefficients and winding conditions, aiming to quickly obtain POD modal coefficients through winding conditions, thereby skipping the complex nonlinear calculations of the reduced order model and efficiently reconstructing the winding temperature field. Related examples show that the method has good computational accuracy and efficiency. Under 50 test conditions, compared with full-order calculations, the error does not exceed 2.5 K, and the total calculation time is only 1.45 s. Finally, a temperature rise test platform is built based on the 110 kV transformer winding. The test results show that the average calculation error of the reduced-order calculation results does not exceed 2 K compared to the test results, and the single-step calculation time is only 0.03 s. Compared with the full-order calculation of the same scale, the calculation efficiency is significantly improved.