Model Selection of Canonical Particle Swarm Optimizer by EPSO: Meta-Optimization

摘要

We have proposed Evolutionary Particle Swarm Optimization, EPSO, which can estimate PSO models for efficiently solving various optimization problems. It provides a new paradigm of meta-optimization for model selection to swarm intelligence and stochastic optimization. In order to inspect its coverage, in this paper we propose to apply EPSO to Canonical Particle Swarm Optimizer (CPSO) for systematically estimating a proper parameter set in CPSO. A crucial idea here is to adopt a temporally cumulative fitness of the best particle as a swarm representative for evaluating the performance of CPSO model. To demonstrate the effectiveness of the proposal, computer experiments on a suite of benchmark problems are carried out. We analysis the characteristics of the obtained CPSO models, and compare the search performance with the results of the original CPSO and RGA/E.