An Efficient Hybrid Sampling Method for Neural Network-Based Microwave Component Modeling and Optimization

Abstract

In this letter, we propose an efficient hybrid sampling method for microwave component modeling and optimization. The sampling method adaptively chooses samples from global and local samples to form a data set. The local samples are obtained using a greedy-like sampling method to exploit potential optimal solutions. The global samples are chosen using random sampling with minimum distance rejection to ensure the uniformity of the samples in the design space. The obtained data set is used to establish a surrogate model using the artificial neural networks (ANNs), and the optimal design parameters are obtained by optimizing the ANN model. A bandstop microstrip filter is taken as an example to verify the performance of the sampling method. The results show that the ANN model based on the proposed method achieves better modeling performance and yields better optimal design than the ANN model based on conventional sampling methods.