A Time Domain Neural Network Method for Scattering of Complex Scatterers

To establish an efficient and high accuracy neural network-based time-domain electromagnetic calculation method, a supervised learning method based on a long short-term memory (LSTM) network is proposed in this work to obtain the time-domain electromagnetic response of scatterers. Based on the propagation direction of electromagnetic waves, the permittivity distribution of scatterers is input into each LSTM cell in turn in this method. To validate the proposed method, the complex scatterers with irregular shape and inhomogeneous permittivity are considered. After training, the results show that the LSTM network can well complete the prediction of time-domain scattering response with high accuracy in the calculation of two-dimensional complex scatterers and a relative error is only 1.3%, while in terms of efficiency, it is more than 400 times faster than the finite-difference time-domain (FDTD) method.