A high-performance computing-based multi-island distributed migrating birds optimization algorithm implemented using MPI

Abstract

Metaheuristic algorithms are commonly used optimization techniques for solving highdimensional complex problems. To achieve more successful results, improvements are made on these metaheuristic algorithms. This study aims to enhance the performance of the Migrating Birds Optimization (MBO) Algorithm, a neighborhood-based metaheuristic algorithm. To this end, MBO is implemented on a high-performance computing (HPC) architecture using Open MPI, and the Multi-Island Distributed Migrating Birds Optimization (MIDMBO) Algorithm is developed. MIDMBO is a distributed metaheuristic optimization algorithm that utilizes multiple islands, where successful solutions migrate between islands. The migration process between islands, facilitated by MPI, enables MIDMBO to generate more successful outcomes. Therefore, studies focused on the migration process are of great importance. In this study, the relationship between algorithm performance and variations in fundamental migration parameters such as migration rate, migration interval, and the number of islands is examined. Regression analyses are performed on the results obtained from MIDMBO, and the model that best explains performance variation is identified. Additionally, artificial neural network (ANN) models are constructed, and the degree to which changes in parameter values explain the results is evaluated.