Feature-Assisted Neural Network Surrogate-Based Multiphysics Optimization for Microwave Filters
Jiaping Fu, Jing Jin, Jingxian Yang, Zixin Liu, Jing Qian, Hai Lin
Abstract
This letter proposes a feature-assisted artificial neural network (ANN) surrogate-based multiphysics optimization technique exploiting trust-region algorithm for microwave filters. The proposed technique introduces the feature assistance (FA) into multiphysics optimization for the first time. In this technique, the proposed feature-assisted multiphysics surrogate model is composed of the multiphysics-based-ANN model and feature-parameters-based-ANN model. The feature-parameter-based ANN model is used to assist the multiphysics-response-based ANN model to search for the optimal solution in the optimization process. Furthermore, we derive a new normalized objective function employing feature parameters to assist the multiphysics optimization. Using the normalized objective function, the proposed technique can find the optimal solution more efficiently than the existing surrogate-based multiphysics optimization methods. We use two waveguide filters as examples to validate the proposed technology.