Particle swarm optimization with mutation for high dimensional problems

Abstract

Particle Swarm Optimization (PSO) is an effective algorithm for solving many global optimization problems. Much of the current research, however, focusses on optimizing functions involving relatively few dimensions (typically around 20 to 30, and generally fewer than 100). Recently PSO has been applied to a different class of problems - such as neural network training - that can involve hundreds or thousands of variables. High dimensional problems such as these pose their own unique set of challenges for any optimization algorithm. This chapter investigates the use of PSO for very high dimensional problems. It has been found that combining PSO with some of the concepts from evolutionary algorithms, such as a mutation operator, can in many cases significantly improve the performance of PSO on very high dimensional problems. Further improvements have been found with the addition of a random constriction coefficient. © 2008 Springer-Verlag Berlin Heidelberg.