Temporal Extrapolation Generalization of Proper Orthogonal Decomposition (POD) and Radial Basis Function (RBF) Surrogates for Transient Thermal Fields in Multi-Heat-Source Electronic Devices

Efficient and accurate prediction of transient temperature fields is critical for thermal management of electronic devices with multiple heat sources. In this study, a reduced-order surrogate modeling approach is developed based on proper orthogonal decomposition (POD) and radial basis function (RBF) neural networks. The method maps time-conditioned modal coefficients in a parameter–time space, enabling robust temporal extrapolation beyond the training horizon. A multi-heat-source conduction model typical of electronic packages is used as the application scenario. Numerical experiments demonstrate that the proposed POD–RBF surrogate achieves high predictive accuracy (global MRE < 3%) with significantly reduced computational cost, offering strong potential for real-time thermal monitoring and management in electronic systems.