Thermal current coupling modeling and heat dissipation structure optimization of switchgear in prefabricated substations based on APINN

Abstract

To address the prominent local temperature rise in switchgear installed in prefabricated substations, where the internal structure is compact, heat sources are concentrated, and ventilation paths are constrained, this paper proposes a thermal-fluid coupled modeling and heat dissipation structure optimization method based on an augmented physics-informed neural network (APINN). A three-dimensional thermal-fluid coupled numerical model was established according to the actual structure and operating conditions of the switchgear, and its reliability was verified by a temperature-rise experiment. The continuity equation, momentum conservation equation, energy conservation equation, and boundary conditions were embedded into the neural-network loss function to construct an APINN surrogate model that combines physics constraints with data supervision. Soft domain decomposition, a gating network, parameter sharing, and implicit turbulence modeling were further introduced to improve the prediction accuracy for local hot spots and strongly disturbed flow fields in a complex enclosed space. The experimental and simulation results show that the APINN reconstructed the temperature field with a root mean square error of 0.62 K and a mean absolute error of 0.48 K. The computation time for a single field prediction was 4.81 s, and the maximum reconstruction error was 1.1 K, outperforming standard PINN and Kriging models. A bi-variable coupled optimization of heat dissipation was then conducted using an entropy-weighted comprehensive benefit function. The optimized scheme reduced the maximum temperature rise by 10.9%, with a maximum verification error below 0.3 K. The proposed method provides technical support for heat dissipation structure design and digital operation and maintenance of switchgear in prefabricated substations.