Efficient RCS Prediction of the Conducting Target Based on Physics-Inspired Machine Learning and Experimental Design

Abstract

In this article, we propose a hybrid approach that combines machine learning and experimental design to efficiently and accurately predict the monostatic radar cross section (RCS) of a conducting target versus the incident angle. The approach is called physical optics-inspired support vector regression (POI-SVR). The design of its kernel function is inspired by PO. Uniform design (UD) and uniform design sampling (UDS) are introduced to obtain highly representative training samples. Numerical experiments dealing with simple and complex targets are carried out to evaluate the accuracy and efficiency of the proposed method. The results show that our method can reduce the predictive root-mean-square error (RMSE) by 29.38%-64.78% compared with the alternative methods of combining a Gaussian SVR with the centrically located sampling (CLS), the Latin hypercube sampling (LHS), or the simple random sampling (SRS). Under the same sampling strategies (i.e., UD and UDS), POI-SVR can reduce the predictive RMSE by 11.30%-53.56% compared with the Gaussian SVR. The well-trained POI-SVR can predict the monostatic RCS of the target in any direction within 0.1 s, and in 20 000 directions within 10 s.