Forking genetic algorithms: GAs with search space division schemes

Area of Science:

• Computational Intelligence
• Optimization Algorithms
• Evolutionary Computation

Background:

• Conventional genetic algorithms (GAs) struggle with multimodal optimization problems.
• Multimodal problems require specialized algorithms to explore diverse solution landscapes effectively.

Purpose of the Study:

• To introduce a new forking genetic algorithm (fGA) designed for multimodal optimization.
• To address the limitations of traditional GAs in handling complex search spaces.

Main Methods:

• The forking GA (fGA) divides the search space into subspaces based on population convergence and solution status.
• A multipopulation scheme is employed, with a parent population exploring and child populations exploiting subspaces.
• Two variants are proposed: genotypic fGA (g-fGA) using schema analysis and phenotypic fGA (p-fGA) using neighborhood hypercubes.

Main Results:

• Both g-fGA and p-fGA demonstrated strong performance on complex function optimization problems.
• The proposed fGA variants outperformed conventional genetic algorithms in empirical evaluations.
• The p-fGA showed additional potential utilities that were briefly explored.

Conclusions:

• The forking genetic algorithm (fGA) offers a promising approach for tackling difficult multimodal optimization problems.
• The g-fGA and p-fGA variants provide effective strategies for search space division and exploration/exploitation.
• This research contributes a valuable enhancement to the field of evolutionary computation for optimization.