The analysis and prediction of key information regarding switchgear are crucial for enhancing the operational lifespan and ensuring the stable functioning of these systems. However

existing data-driven inference methods often exhibit weak interpretability and high dependence on data

while simulation-based approaches tend to have lengthy solution times and poor practical validation. To address these issues

this paper proposes a novel method that integrates both strategies and enables simultaneous forward and inverse thermal rise solutions. Initially

we apply mathematical physics methods alongside machine learning techniques to perform a weak form treatment and functional variational analysis of the heat conduction fluctuation equation within the switchgear. Subsequently

we embed this framework into a deep network to construct a physics-informed neural network (PINN) that combines data with prior knowledge. By leveraging objective laws governing heat generation and conduction for error analysis and computation

we achieve an effective deduction of thermal field distribution in switchgear as well as inversion of fault heat sources. In terms of forward solving the three-dimensional thermal field distribution in switchgear

we introduce a dual-physics information neural network (DPINN) model. This approach reduces the maximum temperature rise error from 6.8% to 0.8%. For unknown heat source inversion

we propose an iterative method where unknown coefficients synchronize with network updates. The maximum error for dual-parameter heat source parameters remains within 2.5%.The results indicate that our DPINN-based network model demonstrates significant advantages in solving both forward and inverse problems related to three-dimensional temperature rise distributions in switchgear. This has meaningful implications for perceiving

analyzing

and diagnosing temperature rise characteristics during operation.