A coevolutionary memetic particle swarm optimizer

Abstract

This paper presents a coevolutionary memetic particle swarm optimizer (CMPSO) for the global optimization of numerical functions. CMPSO simplifies the update rules of the global evolution and utilizes five different effective local search strategies for individual improvement. The combination of the local search strategy and its corresponding computational budget is defined as coevolutionary meme (CM). CMPSO co-evolves both CMs and a single particle position recording the historical best solution that is optimized by the CMs in each generation. The experimental results on 7 unimodal and 22 multimodal benchmark functions demonstrate that CMPSO obtains better performance than other representative state-of-the-art PSO variances. Particularly, CMPSO is shown to have higher convergence speed. © 2012 Springer-Verlag.