Abstract
This paper explores the feasibility of using machine learning(ML) to address time-domain electromagnetic forward and inverse problems. A novel ML-AH FDTD solution method is proposed, which utilizes datasets generated by the AH FDTD method to train the neural network model, and then uses this trained model for prediction. Both the prediction of the target scattering field and the inversion of the target can achieve good results. Moreover, in the task of predicting the locations of metal tunnels, the frequency-domain convolutional neural network(FD-CNN) performs better than Galerkin Transformer(GT), and simulation of this case using AH FDTD method, the required dataset can be obtained in about 2h, whereas FDTD takes nearly 41h.
Funding Information
  • National Key Laboratory Foundation of China (JCKYS2023LD5)
  • National Natural Science Foundation of China (61801217)
  • Natural Science Foundation of Jiangsu Province (BK20180422)