A reduced-order model (ROM) for the temperature field based on time-space proper orthogonal decomposition (POD) is presented to improve the computational efficiency of transient temperature rise in oil-immersed power transformers with a complete oil natural convection cooling loop. This approach utilises data obtained from a full-order simulation model operating under the transformer's rated conditions as input. This paper first examines the reduction characteristics of the proper orthogonal decomposition (POD) method and subsequently introduces the time-space POD framework. The time POD method is used to select a robust snapshot set

which is subsequently integrated with space POD and Galerkin methods to establish a rapid computation model for the transient temperature field prediction of the transformer under various loading conditions. Compared to traditional finite element models

this method can generate a highly accurate computational model with a small number of snapshot samples

significantly reducing the computation time required. The presented examples demonstrate the exceptional computational accuracy and efficiency of this method. Therefore

this approach offers a workable solution for the digital twin application of transformer temperature assessment by attaining second-level computing efficiency for the entire transformer temperature field.