A Domain-adaptive Physics-informed Neural Network for Inverse Problems of Maxwell's Equations in Heterogeneous Media

• DOI: 10.48550/arxiv.2308.06436
• Publication Date: 2023-01-01
• Authors: S. Piao; Hong Gu; Aina Wang; Pan Qin
• Keywords: Electromagnetism; Electromagnetic field solver; Electromagnetics; Lorentz force

Maxwell's equations are a collection of coupled partial differential (PDEs) that, together with the Lorentz force law, constitute basis classical electromagnetism and electric circuits. Effectively solving is crucial in various fields, like electromagnetic scattering antenna design optimization. Physics-informed neural networks (PINNs) have shown powerful ability PDEs. However, PINNs still struggle to solve heterogeneous media. To this end, we propose domain-adaptive PINN (da-PINN) inverse problems First, location parameter media interface decompose whole domain into several sub-domains. Furthermore, conditions incorporated loss function improve prediction performance near interface. Then, training strategy for da-PINN. Finally, effectiveness da-PINN verified two case studies.