Cross-generation Elites Guided Particle Swarm Optimization for large scale optimization

Abstract

Elites have been widely used in many evolutionary algorithms. However, only elites in current generation are utilized to guide the learning/updating of particles/individuals in existing algorithms. Usually, elites in different generations are different and elites in the past generations may contain experienced knowledge and thus may be helpful for guiding particles/individuals to promising areas. Inspired from this, we propose a Cross-generation Elites Guided Particle Swarm Optimizer in this paper. Specifically, the swarm in current generation is divided into two separate sets: the elite set containing the top best particles and the non-elite set consisting of the rest particles. Since these elite particles are the most promising ones in the current generation, we remain these elites unchanged and let them directly enter next generation. Then the rest non-elite particles are updated through learning from elites in both the current generation and the last generation. Through this, a potential balance between exploration and exploitation can be achieved. Particularly, the proposed algorithm is applied to deal with large scale optimization, which is very challenging and difficult and has received a lot of attention in recent years. Extensive experiments are conducted on two sets of large scale benchmark functions and experimental results verify the competitive effectiveness and efficiency of the proposed algorithm in comparison with several state-of-the-art large scale evolutionary algorithms.