Electromagnetic Modeling Using an FDTD-Equivalent Recurrent Convolution Neural Network: Accurate computing on a deep learning framework
Liangshuai Guo, Maokun Li, Shenheng Xu, Fan Yang, Li Liu
Abstract
In this study, a recurrent convolutional neural network (RCNN) is designed for full-wave electromagnetic (EM) modeling. This network is equivalent to the finite difference time domain (FDTD) method. The convolutional kernel can describe the finite difference operator, and the recurrent neural network (RNN) provides a framework for the time-marching scheme in FDTD. The network weights are derived from the FDTD formulation, and the training process is not needed. Therefore, this FDTD-RCNN can rigorously solve a given EM modeling problem as an FDTD solver does.
Topics & Concepts
Finite-difference time-domain methodConvolution (computer science)Computer scienceSolverKernel (algebra)Convolutional neural networkArtificial neural networkRecurrent neural networkFinite difference methodComputational electromagneticsDeep learningAlgorithmArtificial intelligenceElectromagnetic fieldMathematicsPhysicsMathematical analysisDiscrete mathematicsOpticsQuantum mechanicsProgramming languageElectromagnetic Simulation and Numerical MethodsGeophysical Methods and ApplicationsSoil Moisture and Remote Sensing