Application of artificial neural networks for electromagnetic modeling and computational electromagnetics

This paper presents an overview of emerging artificial neural network (ANN) techniques and applications for electromagnetic (EM) simulation and design. Accurate time domain EM modeling using recurrent neural networks (RNNs) is reviewed. Advanced robust training algorithm combining particle swarm optimization (PSO) and quasi-Newton method is described through frequency domain EM modeling, showing its ability to avoid ANN training being trapped in local minima to obtain accurate models. ANN applications in computational electromagnetics are also discussed. Great efficiency can be achieved by using ANNs to approximate the computationally intensive calculations in solving Maxwell equations using method of moments (MoM). As illustrated in examples, these ANN-based techniques are capable of fast and accurate EM modeling and MoM computation, and useful for efficient EM based design.