Multipath Machine Learning Assisted Optimization and Its Application for Antenna Design

A multipath machine learning assisted optimization (MP-MLAO) method is proposed for antenna applications. In practical antenna design tasks, the modulation of the balance between exploration and exploitation of the conventional MLAO method always relies on experience of designers, which largely impacts the performance and robustness of the algorithm. A multipath strategy is introduced based on the implementation of multiple lower confidence bound constants to enhance the robustness of the MLAO algorithm. Moreover, with the aid of variable-fidelity-based modeling, the increase of the computational burden is largely avoided. Compared with the conventional MLAO method, the proposed MP-MLAO achieves obvious performance improvement, which is validated by applying it to design a practical cavity-backed slot antenna.