Elektromanyetik saçılım problemlerinin verimli çözümü için özgün hiyerarşik makine-öğrenmesi-tabanlı method

In this thesis, we propose a novel hierarchical machine-learning-based method to solve electromagnetic scattering problems from perfectly electrical conductor objects. The proposed method solves the discretized surface integral equation and uses a tree structure inspired by the multilevel fast multipole algorithm (MLFMA) to provide a hierarchical multilevel scheme with a controllable error. The novelty of the method comes from the evaluation of the matrix-vector- product for the far-zone interactions based on the one-box-buffer scheme. These interactions are computed in a group-wise manner by approximating the box-to-box free space dyadic Green's function for a given geometry with the help of the artificial neural networks. Therefore, once the neural network models to estimate the free-space Green's function are trained for a certain geometry, they become applicable to all arbitrary scatterers with the same electrical size for any excitation. The preliminary results show that the algorithm provides very accurate results and has a comparable complexity with the available state-of-the-art methods, such as MLFMA. Additionally, the proposed method does not exhibit the well-known low-frequency breakdown problem of MLFMA, and its computational efficiency can be greatly improved by parallelization.