Variable resolution machine learning optimization of antennas using global sensitivity analysis.

Area of Science:

• Electromagnetic engineering
• Computational electromagnetics
• Machine learning applications

Background:

• Global optimization in antenna design faces high computational costs with traditional methods.
• Electromagnetic (EM) simulations are computationally intensive, limiting design iterations.
• Existing surrogate-assisted methods struggle with high-dimensional and nonlinear systems.

Purpose of the Study:

• To develop an innovative, computationally efficient global antenna optimization technique.
• To integrate machine learning with advanced optimization strategies.
• To reduce the reliance on extensive high-fidelity EM simulations.

Main Methods:

• Iteratively refined kriging surrogates combined with particle swarm optimization.
• Fast global sensitivity analysis to establish a reduced-dimensionality search region.
• Variable-resolution electromagnetic (EM) simulations for global search and final tuning.

Main Results:

• The proposed method achieves accurate behavioral models with limited training data.
• Significant reduction in computational costs (CPU time) for antenna optimization.
• Demonstrated superiority over benchmark optimization techniques in comprehensive verification.

Conclusions:

• The developed machine learning framework offers a reliable and efficient approach to global antenna optimization.
• The integration of sensitivity analysis and variable-resolution simulations enhances search reliability and design quality.
• This technique provides substantial computational savings, making complex antenna design more accessible.