Collaboration and Competition Process: A Multi-Teams and Genetic Algorithm Hybrid Approach

The hybridization of genetic algorithms and the simplex method have been proven in literature as useful and promising in optimizations. Therefore, this paper proposes a multi-teams genetic-algorithm (MT-GA) hybrid developed toward extending the previous simplex-GA hybrids. The approach utilizes the simplex method as a united team and multi-teams collaboration and also competition search process in conjunction with the GAs. It is designed such that it has multi-teams with self-evolution (parallel applications of the simplex method), multi-teams communication and even mutual stimulation, and multi-teams survival competition as well as non-elite team breakup for individual relearning (with GAs) and re-forming the new teams. The extension of multi-teams GA thus provides the advantages and as previous simplex-GAs has been proved to outperform a number of other approaches. The experiments in this research show that the MT-GA generally outperforms the existing simplex-GAs for the indices of convergence rate (CPU time required), efficiency (number of function evaluations), and effectiveness (accuracy). Also, a further functional experiment of the MT-GA shows that the MT-GA can be a useful improved algorithm for the function optimization problems. 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 DOI: 10.4018/jalr.2010070107 International Journal of Artificial Life Research, 1(3), 62-90, July-September 2010 63 Copyright © 2010, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited. 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. Therefore, hybridizations of techniques may not only help a solution process prevent the local-optima premature convergence (Yang & Douglas, 1998; Chelouah & Siarry, 2003), but also can assist the solution procedure in more effectively and efficiently locating the global optima. Therefore, inspired by these developments, especially the hybridization of GAs, subpopulation or niche GAs and also the simplex-GA hybrids which have been proven that outperform a number of other approaches, this paper further proposes an extension as the multi-teams genetic algorithm (MT-GA) hybrid that extends the previous simplex-GAs. Also, the present development is due to that existing simplex-GAs show further great improving possibilities; e.g., single-simplex GAs (e.g., Renders & Bersini, 1994; Yen et al. (1998) simplex-GA (Yen et al., 1998); etc.) where the simplex method is only used as an additional crossover, and multi-simplex GAs (e.g., Wei & Zhao, 2005; Hedar & Fukushima, 2003; 27 more pages are available in the full version of this document, which may be purchased using the "Add to Cart" button on the product's webpage: www.igi-global.com/article/collaboration-competitionprocess/46030?camid=4v1 This title is available in InfoSci-Journals, InfoSci-Journal Disciplines Medicine, Healthcare, and Life Science. Recommend this product to your librarian: www.igi-global.com/e-resources/libraryrecommendation/?id=2