Automated Iterative Partitioning for Cooperatively Coevolving Particle Swarms in Large Scale Optimization

Abstract

Particle Swarm Optimization (PSO) is a relatively recent meta-heuristic inspired by the swarming or collaborative behaviour of biological populations. It is known by its capacity of obtaining important fitness improvements on a short period of time. A cooperative version named CPSO has been used to deal with high dimensional search spaces and CCPSO2 is one of its variants that has achieved high performances in large scale optimization problems (above 500 dimensions). This paper proposes an Iterative Partitioning (IP) method for CCPSO2 that takes advantage of the CCPSO2 characteristics. The resulting approach, named CCPSO2-IP, also joins some well known good practices into one single algorithm. Boost functions are included to fine tune the search steps. The competition benchmark CEC13 for large scale global optimization (LSGO) is used to validate the proposed method. Results show that the IP-based method outperforms the standard CCPSO2 and the single swarm PSO, where the exponential boost function presents the highest performance.