Article Preview
Top1. Introduction
Since their inceptions genetic algorithms (GAs) (Holland, 1975) and more broadly evolutionary algorithms (e.g., see Baker, 1985; Deb et al., 2002; Hansen & Ostermeier, 2001; Schmitt, 2001) have become the most utilized techniques for optimizations. They have received great attention from researchers and design engineers from almost all areas, for their ability of direct search by function values only in a mimicked nature evolution process and attainability of global optima in intricate spaces. Also, among the techniques, GAs as the present paper concerns have attracted a great number of researchers providing a great number of extensive works in developing it (e.g., see Goldberg, 1989; Goldberg & Deb, 1991; Janikow & Michalewicz, 1991; Michalewicz, 1992; Mitchell, 1996; Schmitt, 2001; and reviews therein). Also in these developments, a further direction exists, as also concerned here, the hybridization of GAs and other algorithms. A number of researches have evidenced that the hybridization of GAs’ ability in global search with local techniques for their ability of accuracy in the search may enhance the techniques’ convergence rates and search resolution. Thus, as reported the hybridizations have included that with a GA and such techniques as a local neighborhood or hill-climbing technique (e.g., see Ishibuchi et al., 1994; Renders & Bersini, 1994; Ishibuchi & Murata, 1998; Hart et al., 1998; Lin et al., 1998), simplex method (Renders & Bersini, 1994; Yang & Douglas, 1998; Yen et al., 1998; Musil et al., 1999; Hedar & Fukushima, 2003; Chelouah & Siarry, 2003; Ustun et al., 2005; Wei & Zhao, 2005), Levenberg-Marquardt optimizer (e.g., see Park & Froment, 1998), tabu search (e.g., see Chang & Lo, 2001), simulated annealing (e.g., see Hwang & He, 2006), etc. among others, and also a number of others hybridized and reviewed in these articles. In addition, a classification (Yen et al., 1998) has been reported and divides the hybrids of GAs into these categories, herein with added references: (i) pipeline hybrids use the GAs and another technique sequentially (e.g., the G-bit improvement on GA (Georgiou, 2007; Park & Froment, 1998)). (ii) Asynchronous hybrids utilize the shared population to allow for the GAs and another technique to cooperate asynchronously and proceed independently (e.g., the asynchronous teams hybridizing the GA and Newton method (de Souza & Talukdar, 1991)). (iii) Hierarchical hybrids utilize the GAs and another technique in two levels of a problem (e.g., G/SPLINES (Rogers, 1991) hybridizing the GA and multivariate adaptive regression spline). And (iv) other hybrids introduce the additional operators for GAs (e.g., simplex method as an additional crossover (Renders & Bersini, 1994), Yen et al.’s simplex-GA hybrid (Yen et al., 1998), and other simplex-GAs). Meanwhile, in GA developments, on the other hand, also another direction exists, which is termed niche techniques. Niche GAs are those that are able to identify subpopulations (subsystems, or niches) in GAs explicitly/implicitly and develop such techniques as clearing, fitness sharing, clustering-based, crowding, speciation tree, and multi-population GAs (e.g., see Cedeño & Vemuri, 1999; El Imrani et al., 2000; Goldberg & Richardson, 1987; Gurfil & Kasdin, 2002; Mahfoud, 1994; Säreni & Krähenbühl, 1998; Siarry et al., 2002; Wang & Wu, 1999; and reviews therein). The techniques have the origin in preserving the diversity of genetic searches for preventing the local-optima premature convergence and also have evolved to attaining multi-solutions for a multimodal problem. More of these techniques therefore may be reviewed in a later section, when laying further research directions from the current.