Deception, blindness and disorientation in particle swarm optimization applied to noisy problems

Particle swarm optimization (PSO) is a population-based algorithm designed to find good solutions to optimization problems. However, if the problems are subject to noise, the quality of its results significantly deteriorates. Previous works have addressed such a deterioration by developing noise mitigation mechanisms to target specific issues such as handling the inaccurate memories of the particles and aiding the particles to correctly select their neighborhood best solutions. However, in spite of the improvements achieved, it still remains uncertain the extent to which these issues affect the particles, and the underlying reasons for the deterioration of the quality of the results. In this article, we formally define deception, blindness and disorientation as the conditions responsible for such a deterioration, and we develop a set of population statistics to measure the extent to which these conditions affect the particles throughout the search process. The population statistics are computed for the regular PSO algorithm and for PSO with equal resampling (PSO-ER) on 20 large-scale benchmark functions subject to different levels of multiplicative Gaussian noise. The key findings that we reveal with the population statistics on optimization problems subject to noise are the following: (a) the quality of the results significantly deteriorates as particles suffer from large proportions of deception and blindness; (b) the presence of deception, blindness and disorientation, and their effects on the quality of results, makes the performance of the swarms sensitive to optimization problems subject to noise; (c) the incorporation of resampling methods into PSO significantly improves the quality of the results; and (d) it is better to first address the conditions of blindness and disorientation before addressing deception.