A Maxwell's Equations Based Deep Learning Method for Time Domain Electromagnetic Simulations
Pan Zhang, Yanyan Hu, Yuchen Jin, Shaogui Deng, Xuqing Wu, Jiefu Chen
Abstract
In this study, we discuss an unsupervised deep learning approach for time-domain electromagnetic simulations. Our method, based on physics informed neural network, encodes initial conditions, boundary conditions as well as governing equations as the constraints when training the network, turning an electromagnetic simulation problem into an optimization process. High prediction accuracy of the electromagnetic fields, without discretization or interpolation in space or in time, can be achieved with limited numbers of layers and neurons in each layer of the neural network. We analyze three aspects that influence the performance of the deep learning method. First, we discuss the applicability of this network in the homogeneous media and the relative error can be less than 0.1%. Then, we analyze the relationship between accuracy and network architecture in training. Testing results show that increasing the number of hidden layers and neurons per layer can improve prediction accuracy. Finally, we study numerical examples of the inhomogeneous medium.