Novelty search in particle swarm optimization

This paper presents a novel approach to implementing the Novelty search technique (introduced by Kenneth O. Stanley) into the Particle Swarm optimization algorithm (PSO). PSO is well-known for its impaired ability to operate in multidimensional spaces due to its inclination towards premature convergence and possible stagnation. This presented research aims to try various implementations of Novelty Search that could remove this inability and enhance the PSO algorithm. In total, we present five different modifications. The CEC 2020 single objective bound-constrained optimization benchmark testbed was used to evaluate the different Novelty Search-based modifications of the algorithm. All results were compared and tested for statistical significance against the original variant of PSO using the Friedman rank test. This work aims to increase understanding of implementing new approaches for population dynamics control, which are not driven purely by a gradient, and inspire other researchers working with different evolutionary computation methods