Solving Two-Dimensional Scattering From Multiple Dielectric Cylinders by Artificial Neural Network Accelerated Numerical Green's Function

Abstract

A number of real-life applications involve the determination of scattered field of an unchanged large object and several varying small scatterers in its vicinity. Integral equation methods based on free-space Green's function (FSGF) are usually applied to solve this problem. In this letter, we propose an artificial neural network (ANN) accelerated numerical Green's function (NGF) to reduce the computational cost of conventional methods. For explicit demonstration, 2-D scattering from infinite cylinders is considered. Since the unchanged large object is considered as part of the background, its scattering is included in NGF. In our approach, its scattering is extracted via an FSGF scheme in order to exploit the regression ability of ANN efficiently. Then a dataset in possession of the scattering feature is constructed. By feeding ANN with the dataset, ANN can be trained as a good representation of NGF, where only a small number of ANN parameters are computed and stored. Since the dataset only contains part of all the possible field-source pairs in the computation region, the calculation of NGF can be accelerated. Once the ANN accelerated NGF is obtained, the scattering analysis can be easily conducted with a small numerical system where the unknowns are only associated with small scatters. The combination of ANN and NGF assists us to construct a framework that can save the CPU time and memory usage in both online scattering solver and offline NGF solver.