Application of two-dimensional AWE algorithm in training neural networks

Artificial neural networks (ANN) play a very important role in current microwave fields. Training a neural network model is the key to neural network technique. The conventional methods for generating EM training data, such as method of moments (MoM), are time-consuming when the training parameters are a bit more. In order to aid the training process by reducing the amount of costly and time-consuming sampling cycles, fast algorithms need to be used, such as asymptotic waveform evaluation (AWE). Two-dimensional AWE method can do the extrapolation in two dimensions of variables simultaneously, so it has better performance than one-dimensional AWE method. In this paper, the two-dimensional AWE method and neural network technique are combined, so the process of training neural net model can be completed quickly and accurately. Numerical results demonstrate the accuracy and efficiency of this technique.