AN EFFECT ANALYSIS OF THE PARALLEL MIGRATING BIRDS OPTIMIZATION ALGORITHM PARAMETERS

Abstract

Migration is the process of sending selected solutions from a sub-population to the neighboring sub-population at specified intervals in parallel metaheuristic algorithms (PMAs). Topology, migration rate (MR), migration interval (MI), migration policy and communication model are the factors which characterize the nature of migration. Identification of relationship between migration parameters and an accurate selection of such parameter values increase the performance of PMAs. The number of sub-populations (NS) denotes the number of different populations in which algorithm can perform simultaneous searches. In this study, Migrating Birds Optimization (MBO) Algorithm, no migration performed, was applied for four different NS values. Additionally, Parallel Migrating Birds Optimization (PMBO) Algorithm is executed using five MR values, five MI values and four NS values and obtained fitness values are provided. According to the results, PMBO algorithm outperforms MBO in 99% of case studies. Therefore, the contribution of migration to the performance of the algorithm is evidently demonstrated. Furthermore, the values obtained during the iterations are shown on graph to investigate the effect of MI and MR changes on search performance of algorithms. As MI decreases, it is confirmed that the algorithm produces good results in early steps of iterations, making faster searches. MR has a greater effect on performance if MI is kept low. If MI increases, the changes in MR have less affect. Additionally, the effect of MI, MR, NS values and their correlation on fitness value is analyzed with analysis of variance (ANOVA). According to the analysis, MI is identified to be the most significant factor. The least significant factor is NS. Combinations of such parameters are analyzed and it was shown that MR*MI combination has the most significant effect on performance.