Efficient Antenna Modeling and Optimization Using Multifidelity Stacked Neural Network

Abstract

In this work, a multifidelity stacked neural network (MFSNN) is proposed to construct surrogate model for antenna modeling and optimization. The stacked neural network consists of a low-fidelity (LF) network, a linear high-fidelity (HF) network, and a nonlinear HF network. By learning the prior from sufficient computationally cheap LF data, the MFSNN has significantly reduced the requirement of computationally expansive HF data. The correlation between LF and HF models can be learned adaptively and accurately by decomposing the correlation into linear component and nonlinear component. The feasibility of the approach is validated by two antenna structures, which shows that the MFSNN-based surrogation model can make predictions for broad ranges of input parameters with satisfactory accuracy. Then, the surrogate model is directly applied in the particle swarm optimization (PSO) framework to replace the full-wave simulation and accelerate antenna optimization procedure.