Título: Adaptive Radiation as an Autotuning Strategy for Genetic Algorithms on Dynamic Problems
Autores: Heictor Costa , Roberto Oliveira
Resumo: As optimization processes have become more complex for embracing problems of different characteristics, it is necessary to have multiple adequate algorithms to tackle each of such problems. Bio-inspired evolutionary algorithms are suitable solutions for these situations, but they require re-tuning when working with different systems. Autotuning is a popular strategy to increase the adaptability of optimization algorithms. Adaptive Radiation (AR) is a phenomenon in nature that optimizes a population by diversity increase and niche specialization through intense mutation. This research aimed to insert this effect into the Genetic Algorithm (GA) workflow as a biological-inspired autotuning method, creating a new model called Genetic Algorithm with Adaptive Radiation (GAAR). The implementation of the AR component resulted in consistent and improved results on multiple benchmark functions from the CEC2019 challenge. The GAAR only changes the value of the AR component, which is enough to make this model achieve the best results in 57% of the tests and the worst results in 0% of the tests, while the Adaptive Particle Swarm Optimization (APSO) presented 39% and 12% of the best and worst results, respectively
Palavras-chave: Adaptive Radiation, Autotuning, Dynamic Problems
Páginas: 9
Código DOI: 10.21528/CBIC2023-050
Artigo em pdf: CBIC_2023_paper050.pdf
Arquivo BibTeX: CBIC_2023_050.bib