Efficient microwave filter design using differential evolution algorithm with chain local search in MOEA/D framework

Abstract

A hybrid multi-objective differential evolution algorithm based on decomposition with a chain local search strategy (HMODEA/D-CLS) is proposed for electromagnetic (EM) optimization in microwave filter design. This algorithm uses the individual selection based on diversity and the chain local search strategy to approach a local extreme point swiftly. The adaptive allocation strategy of computing resources is used to calculate the time for global optimization. Additionally, a one-dimensional convolutional autoencoder (1D-CAE) as a surrogate model to accelerate the design process. A wideband bandpass filter is designed to verify the algorithm. Compared to the MOEA/D algorithm, the proposed algorithm improves optimization time by approximately sevenfold.