Empirical Study of Particle Swarm Optimization Mutation Operators

Particle Swarm Optimization (PSO) is a global optimization algorithm for real valued problems. One of the known positive traits of the algorithm is fast convergence. While this is considered a good thing because it means the solutions are found faster it can lead to stagnation at a local minimum. There are several strategies to circumvent this. One of them is the use of mutation operators to increase diversity in the swarm. Mutation operators, sometimes called turbulence or craziness operators, apply changes to solutions outside the scope of the PSO update rules. Several different such operators are proposed in the literature mostly based on existing approaches in evolutionary algorithms. However, it is impossible to say which mutation operator to use in which case and why. There is also some controversy whether mutation is necessary at all. We use an algorithm that generates test functions with given properties number of local minima, dimensionality, global minimizer attraction region placement and attempt to explore the relationship between function properties and mutation operator choice. An empirical study of the operators proposed in literature is given with extensive experimental results. Recommendations for choosing appropriate mutation operators are proposed.