Machine learning for microwave optimization using simplex surrogates, dual-resolution computational models and local...

Area of Science:

• Microwave Engineering
• Computational Electromagnetics
• Optimization Algorithms

Background:

• Electromagnetic (EM) analysis is crucial for reliable microwave design.
• Parameter tuning at the EM level is computationally expensive, especially for global optimization tasks.
• Conventional methods struggle with the high costs associated with complex microwave structure optimization.

Purpose of the Study:

• To introduce a novel, fast, and computationally efficient approach for the global optimization of microwave structures.
• To reduce the computational burden of parameter tuning in microwave design.
• To enable reliable and cost-effective optimization for complex designs like metasurfaces.

Main Methods:

• Processing operating parameters instead of full frequency characteristics.
• Utilizing simplex-based regressors to regularize the objective function.
• Employing dual-fidelity EM simulations and restricted sensitivity updates.

Main Results:

• The novel algorithm significantly accelerates the identification of optimum designs.
• Achieved remarkable computational efficiency, averaging fewer than 50 EM simulations.
• Demonstrated superiority over benchmark approaches in optimizing microstrip components.

Conclusions:

• The proposed method offers a computationally efficient and reliable solution for microwave structure optimization.
• The approach simplifies implementation and handling with minimal tunable parameters.
• This technique facilitates faster and more cost-effective design of complex microwave components.