基于BiGRU的FDTD高效精确场预测算法OA

The Finite-Difference Time-Domain(FDTD)method is an important technique in computational electromagnetics and can handle electromagnetic problems involving different shapes,materials,and boundary conditions.However,its computational cost and the constraints imposed by the Courant-Friedrichs-Lewy(CFL)condition limit simulation speed.A method based on a Bidirectional Gated Recurrent Unit(BiGRU)is proposed to predict the dynamic evolution of electromagnetic fields.By training the model to predict field increments between adjacent time steps,numerical stability and learning efficiency are significantly enhanced.A key innovation lies in adopting a highly simplified set of field components as input,which substantially reduces model complexity.After training on 2D TMz FDTD simulation data incorporating sinusoidal wave sources and absorbing boundary conditions,the model successfully achieved multi-step field evolution prediction.Experimental results show that the BiGRU-based prediction model can accurately reproduce FDTD solutions,with relative field errors below-30 dB over 60 prediction steps,offering a new strategy for leveraging neural networks to accelerate electromagnetic simulations.