Fast Adaptive Modeling of Frequency-Domain RCS Responses by Gaussian Process Regression
Lixin Guo, Donghai Xiao, Muyu Hou, Yanchun Zuo, Wei Liu
Abstract
A fast adaptive surrogate modeling technique for analyzing the target's radar cross section (RCS) response versus frequency is proposed based on the Gaussian process regression (GPR). Specifically, an iterative process of modeling and sampling, which seeks the representative points (such as extreme points and inflection points) of the RCS curve, is presented to adaptively determine the required samples and progressively improve the modeling fidelity of the GPR. Validation experiments based on two exemplary targets are performed. Compared with the traditional GPR-based surrogate modeling technique employing a one-shot sampling strategy, the proposed adaptive GPR-based surrogate modeling technique further reduces the computational workload (more than 30%) while maintaining high accuracy.