Abstract: To improve the computational efficiency of the winding temperature in oil-immersed power transformers, this paper proposes a rapid calculation method that integrates discrete node temperatures with the non-intrusive proper orthogonal decomposition (POD) method. First, the POD method is used to establish a reduced-order model for winding temperature calculation, and the reduced-order modes reflecting the characteristics of the winding temperature field distribution are obtained. Next, several discrete nodes within the field are selected to establish a response surface model correlating node temperatures with winding operating conditions. Finally, a mathematical relationship between discrete node temperatures and the entire winding temperature field is constructed using the reduced-order calculation model, enabling rapid inversion from winding operating conditions to discrete node temperatures and subsequently to the entire field temperature. A two-dimensional heat transfer model of a 110 kV oil-immersed transformer winding is analyzed. The results show that the maximum average absolute error of the temperature field does not exceed 0.49 K, the maximum average relative error does not exceed 2.69%, and the maximum error in hot-spot temperature does not exceed 1.72 K. Furthermore, the calculation time for the winding temperature distribution is only 0.78 seconds, representing nearly a thousandfold improvement in efficiency compared to full-order calculations, fully demonstrating the accuracy and high efficiency of the proposed algorithm, and highlighting its significance for establishing high-precision digital twin models of transformers. The study can provide a new approach for the rapid perception and proactive warning of physical entities through digital twin models.