Island flower pollination algorithm for global optimization
Mohammed Azmi Al-Betar,Mohammed A. Awadallah,Iyad Abu Doush,Abdelaziz I. Hammouri,Majdi Mafarja,Zaid Abdi Alkareem Alyasseri
DOI: https://doi.org/10.1007/s11227-019-02776-y
IF: 3.3
2019-03-11
The Journal of Supercomputing
Abstract:Flower pollination algorithm (FPA) is a recent swarm-based evolutionary algorithm that was inspired by the biological evolution of pollination of the flowers. It deals with a panmictic population of pollens (or solutions) at each generation, using global and local pollination operators, to improve the whole population at once. Like other evolutionary algorithms, FPA has a chronic shortcoming that lies in its inability to maturely converge. This is conventionally known as a premature convergence where the diversity of the population is loosed and thus the search is stagnated. Island model is one of the successful structured population techniques that were utilized in the theoretical characteristics of several evolutionary-based algorithms. In this model, the population is divided into a set of islands. The knowledge is distributed among those islands using a migration process that is controlled by migration rate, topology, frequency, and policy. In this paper, the island model is utilized in the evolution process of FPA to control diversity. The proposed approach is called IsFPA. The ability of IsFPA in maintaining the diversity during the search process, and in producing impressive results, can be interpreted by utilizing the island model in the FPA optimization framework. To assess the efficiency of IsFPA, 23 benchmark functions with various sizes and complexities were used. The best parameter configurations of IsFPA were investigated and analyzed. Comparing the results of IsFPA with those of state-of-the-art methods which are FPA, genetic algorithm (GA), particle swarm optimization (PSO), gravitational search algorithm (GSA), multi-verse optimizer (MVO), island bat algorithm (iBA), and island harmony search (iHS), the comparison results show that the IsFPA is able to control the diversity and improves the outcomes where IsFPA is ranked first followed by FPA, iBA, iHS, GSA, MVO, GA, PSO, respectively, based on the Friedman test with Holm and Hochberg as post hoc statistical test.