Skip to content
Open access

Temperature-field reconstruction and heat-source inversion estimation method for high-voltage switchgear based reduced-order PINN surrogate model

Oct 2026 · Engineering Research Express · Vol 8 · 0 citations · 20 references
Physics

Abstract

Localized overheating induced by poor electrical contact, loosened busbar connections, and insulation degradation is a major precursor to failures in high-voltage switchgear. To address the difficulty in directly measuring internal heat sources, a fast three-dimensional temperature-field reconstruction, candidate heat-source identification, and heat-source power inversion method for high-voltage switchgear is proposed based on a reduced-order surrogate model integrating proper orthogonal decomposition (POD) and physics-informed neural networks (PINNs). First, a coupled three-dimensional thermal-flow coupled computational fluid dynamics model is established to generate temperature fields under different heat-source conditions. POD is then applied to extract the dominant thermal modes and their corresponding low-dimensional modal coefficients from the high-dimensional temperature fields. A PINN-based reduced-order surrogate model is subsequently trained to establish the nonlinear mapping from heat-source parameters to POD modal coefficients, while residuals of the steady-state heat conduction equation and boundary-condition constraints are incorporated into the loss function to enforce physical consistency and enhance prediction accuracy. Based on external temperature measurements, an improved grey wolf optimizer is further adopted to identify the active fault heat source from a predefined set of candidate locations and inversely estimate its corresponding heat-source power. Numerical simulations and temperature-rise experiments show that the proposed method accurately reconstructs the three-dimensional temperature distribution and identifies local high-temperature regions. In the temperature-rise experiment, the relative error of heat-source power inversion is only 2.27%, demonstrating that the proposed framework enables rapid and accurate temperature-field reconstruction and heat-source inversion.

Read PDF

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.