DAVMAS-GP: Domain aware variance minimizing Gaussian process regression for complex monostatic RCS prediction

We present our method DAVMAS-GP, a Domain Aware Variance Minimizing Adaptive Sampling Gaussian Process for the prediction of RCS characteristics, including the complex vertically (VV) and horizontally (HH) polarized scattered far-field in both the angular and frequency domains. The method uses Gaussian process regression at its core, but employs the usage of nonstationary kernels, and our adaptive sampler VMAS to attain high accuracy predictions under extremely sparse sampling conditions. We validate our method with an aircraft model which exhibits complex scattering phenomena. Numerical results show that DAVMAS-GP is able to reduce the predictive root-mean-square-error (RMSE) by at least 98.5% compared to traditional methods of combining a Matérn kernel with non-informed Latin Hypercube sampling (LHS). With VMAS, < 1% RMSE is achieved using only 2% of samples. Allowing up to 4% of samples enables < 0.05% RMSE, across all vertical and horizontal complex components.

Name thesis:degree_name

M.S.

Level thesis:degree_level

Thesis

Discipline thesis:degree_discipline

Electrical & Computer Engr

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2024

Author dc:creator

• Jao, Kenneth

Contributors dc:contributor

• Peng, Zhen

• Computational Electromagnetics
• Covariance Function
• Gaussian Process Regression (gpr)
• Surrogate Model
• Nonstationary Kernel
• Variance Minimization
• Adaptive Sampler
• Radar Cross Section (rcs)

Statement dc:rights

• Copyright 2024 Kenneth Jao

Language dc:language

en, eng

Handle dc:identifier

https://hdl.handle.net/2142/125760

OAI record oai:identifier

https://www.ideals.illinois.edu/oai-pmh?verb=GetRecord&metadataPrefix=etdms&identifier=oai%3Awww.ideals.illinois.edu%3A2142%2F125760