Abstract:

To restrain the precocity problem in the artificial bee colony algorithm (ABC),an ensemble learning framework is proposed to discover more useful information that lies in the current population. When an individual produces a candidate, an ensemble best (ebest) solution will be generated by linearly combining all the solutions better than the current solution, and then the candidate will be generated by using corresponding ABC’s search equations, but the global best solution (gbest) term in the search equations will be replaced by the ebest term. The proposed framework provides more promising solutions to guide the evolution and effectively helps ABCs escape the local optima. Experiments show that the novel ensemble learning framework can significantly improve the performance of gbest guided ABCs without increasing their complexity. Moreover, the proposed framework can be utilized to improve the performance of particle swarm optimization and differential evolution variants.

Key words: artificial bee colony, ensemble learning, particle swarm optimization, differential evolution, evolution computation