Electromagnetic Field Prediction in 2D Inhomogeneous Distributions of Permeability with Galerkin Neural Operator
Xiu-Zhen Gong, Zheng‐Yu Huang, Eng Leong Tan, Xue-Zhi Zheng, Yi-Ru Zheng, Han-Yan Duan, Feng Jiang · 2024
The layer normalization scheme in Galerkin neural operator (GNO) allows scaling to propagate to the attention layer, thereby significantly aiding the model in operator learning tasks which involve unnormalized data. This paper explores GNO as an alternative to traditional simulations in solving TEz field in a two-dimensional inhomogeneous distributions of permeability. GNO can learn complex electric field patterns and quickly predict their behavior after training, greatly accelerating the simulation process and improving efficiency. Utilizing GNO helps to gain a deeper understanding of electric field behavior, speeding up the design and optimization process. It helps to discover the impact of dielectric changes on the behavior of the electric field, providing profound insights for waveguide designs. Through the training of GNO, we can also explore the complex laws underlying the behavior of the electric field, bringing new breakthroughs and progress to the research and application in the field of electromagnetics. In this paper, the training data is obtained by simulating the TE mode of the two-dimensional plane waveguide by FDTD.