Linkage Sensitive Particle Swarm Optimization

Abstract

In many optimization problems, the information necessary to search for the optimum is found in the linkages or inter-relationships between problem components or dimensions. Exploiting the information in linkages between problem components can help to improve the quality of the solution obtained and to reduce the computational effort required. Traditional particle swarm optimization (PSO) does not exploit the linkage information inherent in the problem. We develop a variant of particle swarm optimization that uses these linkages by performing more frequent simultaneous updates on strongly linked components. Prior to applying the linkage-sensitive variant of PSO to any optimization problem, it is necessary to obtain the nature of linkages between components specific to the problem. For some problems, the linkages are known beforehand or can be set by inspection. In most cases, however, this is not possible and the problem-specific linkages have to be learned from the data available for the problem under consideration. We show, using experiments conducted on several test problems, that the quality of the solutions obtained is improved by exploiting information held in the inter-dimensional linkages.