Evolutionary Optimization of Dielectric Lens Design through Biased Initialization

Alp İ.

Erzincan Üniversitesi Fen Bilimleri Enstitüsü Dergisi, vol.19, no.2, pp.377-392, 2026 (TRDizin)

Abstract

In high-dimensional dielectric lens designs, randomly generated initial populations often increase the number of full-wave electromagnetic simulations and computational costs due to the presence of low-performance solutions. In this study, a hybrid optimization approach is proposed that combines a statistically guided initial population with a genetic algorithm and simultaneous perturbation stochastic approximation to reduce this problem without restricting the design space. The performance of the proposed approach is evaluated on a fully dielectric planar lens using full-wave electromagnetic simulations based on the two-dimensional finitedifference time-domain solution of Maxwell’s equations. By employing the biased initialization strategy, the average fitness value is improved from below 0.4 to above 0.77. Furthermore, a similarity exceeding 88% and focusing performance close to the diffraction limit are achieved for an ideal point source, while an input–output correlation above 98% is obtained for a more realistic input distribution. These results demonstrate that the proposed approach provides an effective and generalizable optimization strategy for high-dimensional electromagnetic inverse design problems by enabling faster convergence with fewer simulations.