The Use of Physical-informational Neural Networks in the Field of Electromagnetic Field Solving

The current underutilization of existing physics-informed neural networks (PINN) in the domain of electromagnetic fields necessitates a concerted effort to facilitate the gradual transition from finite element methods to PINN. The complexity of simulating electric fields across various media using PINN presents significant challenges, particularly when addressing multiple equations and media in the context of magnetic field analysis, where numerous magnetic field quantities must be resolved. This study employs PINN to achieve the solution of the electric field at a simple boundary, enabling the simulation of electric fields in diverse media by accounting for varying conductivity. Additionally, the research addresses the dynamics of electric field solutions in response to time-varying boundaries. Subsequently, multiple equations are encoded concurrently, and boundary conditions are established to facilitate the simultaneous prediction of multiple magnetic fields. Ultimately, the methodology addresses the resolution of vector magnetic fields in nonlinear materials across multiple media. The accuracy of the proposed approach is corroborated through a comparative analysis with the finite element method.