A Fast Calculation Method Integrating POD-SSA-KELM for Steady-State Temperature Field of Oil-Immersed Transformer Windings

To address the low computational efficiency of steady-state temperature field of oil-immersed power transformer windings, a fast calculation method is proposed, in which the Proper Orthogonal Decomposition (POD) order reduction technique is combined with a Sparrow Search Algorithm (SSA) and optimized Kernel Extreme Learning Machine (KELM) model. First, a full-order numerical model of the steady-state temperature field of the transformer windings is constructed, and the snapshot matrix is generated through parametric simulations. Next, POD is applied to reduce the order of the snapshot matrix, extracting the N dominant modes and their modal coefficients, which cover 99% of the total energy. To overcome the efficiency bottleneck of POD when dealing with strong nonlinear problems, the SSA algorithm is introduced to globally optimize the radial basis kernel function parameters and the regularization coefficient of the KELM, establishing a nonlinear mapping surrogate model between operating conditions and modal coefficients. The temperature field can then be rapidly reconstructed by linearly weighting the modal coefficients and dominant modes. In 30 test conditions, the average absolute error of the temperature field does not exceed 1.51K, with a total computation time of only 0.5642s. Furthermore, a temperature distribution measurement experimental platform for a 110kV transformer winding is built to experimentally validate the proposed method. The results show that the average relative error between the computed results and the measured data is no more than 4.66%, with a computation time of only 0.367s, achieving approximately 21 times faster than Fluent software simulation (which takes 7.86s) and improving the accuracy by about 0.37%. This method significantly enhances computational efficiency while ensuring calculation accuracy, providing efficient and reliable technical support for real-time monitoring and optimal design of transformer winding temperature field.