Bioluminescent Swarm Optimization Algorithm

Evolutionary Computation (EC) is a research area of metaheuristics mainly applied to real-world optimization problems. EC is inspired by biological mechanisms such as reproduction, mutation, recombination, natural selection and collective animal behavior. Two branches of EC can be highlighted: Evolutionary Algorithms (EA) comprising Genetic Algorithms (Goldberg, 1989), Genetic Programming (Koza, 1992), Differential Evolution (Storn & Price, 1997), Harmony Search (Geem et al., 2001), and others; and Swarm Intelligence (SI) comprising Ant Colony Optimization (ACO) (Dorigo & Stutzle, 2004) and Particle Swarm Optimization (PSO) (Kennedy & Eberhart, 2001; Poli et al., 2007) and others. The ACO metaheuristic1 is inspired by the foraging behavior of ants. On the other hand, the PSO metaheuristic2 is motivated by the coordinated movement of fish schools and bird flocks. Both ACO and PSO approaches have been applied successfully in a vast range of problems (Clerc, 2006; Dorigo & Stutzle, 2004). In recent years, new SI algorithms were proposed. They have in common biological inspirations, such as bacterial foraging (Passino, 2002), slime molds life cycle (Monismith & Mayfield, 2008), various bees behaviors (Karaboga & Akay, 2009), cockroaches infestation (Havens et al., 2008), mosquitoes host-seeking (Feng et al., 2009), bats echolocation (Yang, 2010), and fireflies bioluminescense (Krishnanand & Ghose, 2005; 2009; Yang, 2009). This work proposes a new swarm-based evolutionary approach based on the bioluminescent behavior of fireflies, called Bioluminescent Swarm Optimization (BSO) algorithm. The BSO uses two basic characteristics of the Glow-worm Swarm Optimization (GSO) algorithm proposed by (Krishnanand & Ghose, 2005): the luciferin attractant, and the stochastic neighbor selection. However, BSO goes further introducing new features such as: stochastic adaptive step sizing, global optimum attraction, leader movement, and mass extinction. Besides, the proposed algorithm is hybridized with two local search techniques: local unimodal sampling and single-dimension perturbation. All these features makes BSO a powerful algorithm for hard optimization problems. Experiments were done to analyze the sensitivity of the BSO to control parameters. Later, extensive experiments were performed using several benchmark functions with high