Machine Learning Based Multilevel Fast Multipole Algorithm

In this paper, a novel translator calculation method for the multilevel fast multipole algorithm (MLFMA) is proposed based on a machine learning approach. The generalized regression neural networks are introduced to fit the translation function of MLFMA during the procedure of eletromagnetic scattering analysis. Compared to the traditional method, the new method inherits advantages of the generalized regression neural networks (GRNN) and can approximate the translator with high accuracy simultaneously. As an example, a two-level 2D MLFMA for a perfect electrically conductor (PEC) is finally implemented and the translators are reconstructed by using the proposed model. The numerical results validate the effectiveness of the proposed method.