Effect of Fitness Functions on the Performance of Evolutionary Particle Swarm Optimization

Abstract

This paper presents an Evolutionary Particle Swarm Optimization (EPSO) for PSO model selection. It provides a new paradigm of meta-optimization that systematically estimates appropriate values of parameters in PSO model for efficiently solving various optimization problems. In order to further investigate the characteristics, i.e., exploitation and exploration in search, of the PSO model optimized by EPSO, we propose to use two fitness functions in EPSO, which are a temporally cumulative fitness of the best particle and a temporally cumulative fitness of the entire swarm for designing PSO models. Applications of the proposed method to some benchmark optimization problems well demonstrate its effectiveness. Our experimental results indicate that the former fitness function can generate a PSO model with higher fitness, and the latter can generate a PSO model with faster convergence.