Combination of Uniform Design with Support Vector Regression to Assist Analysis of Electromagnetic Scattering from Targets

The analysis of electromagnetic scattering characteristics of complex targets has always been a hot topic in computational electromagnetics, as well as a theoretical and technical support for target recognition and stealth design. However, both numerical methods (like MoM) and approximation methods (like PO) have obvious shortcomings: low efficiency of the former and poor accuracy of the latter. In this paper, we proposed a hybrid approach of uniform design (UD) and support vector regression (SVR) to assist analysis of electromagnetic monostatic scattering from complex targets. The core idea of this approach is to establish an efficient and reliable regression model through effective sampling. Uniform design is adopted to guide sampling backscattering data for its capability of producing samples with high representativeness. For the sake of efficiency, good lattice point with power generator is selected to construct large uniform designs, which can ensure that the obtained uniform design tables have considerable uniformity while greatly reducing the computational complexity. Afterward, with the aid of multilevel fast multipole method (MLFMM), the required datasets are established on the simulation data of a SLICY model and then are utilized to train the support vector regression (SVR) model so as to obtain an efficient and accurate nonlinear regressor. Finally, comparison of the two approaches, one of which is the support vector regression (SVR) combined with uniform design (UD) and the other is that combined with centrally located sampling (CLS), is taken into account, and the result shows that the proposed method can effectively improve the prediction accuracy of regression model and the analysis efficiency of target electromagnetic scattering characteristics. That means the integrated application of experimental design and data mining to computational electromagnetics does make sense.