In recent years, wireless power transmission and energy harvesting have gained significant attention for powering self-sustaining or battery-less devices, such as wireless sensors. These systems require microwave planar circuits that often incorporate nonlinear elements, making their design complex and computationally intensive. To address these challenges, automated design using Genetic Algorithms (GA) has been widely explored. However, the requirement for repetitive electromagnetic field analysis during fitness evaluation remains a major computational bottleneck. This study focuses on the inherent nature of GA, where high population diversity results in numerous individuals with low fitness. Since these unpromising individuals have a low probability of survival, their detailed evaluation can be omitted to save computational resources. We propose an accelerated GA framework that integrates image processing for feature extraction from circuit patterns and a LightGBM-based machine learning model for rapid classification. The proposed LightGBM-assisted GA reduced the total number of fitness calculations by approximately 86% compared to conventional GA-only methods, without sacrificing the quality of the final design.
Akada et al. (Thu,) studied this question.