An Adaptive Velocity Particle Swarm Optimization for high-dimensional function optimization

Researchers have achieved varying levels of successes in proposing different methods to modify the particle´s velocity updating formula for better performance of Particle Swarm Optimization (PSO). Variants of PSO that solved high-dimensional optimization problems up to 1,000 dimensions without losing superiority to its competitor(s) are rare. Meanwhile, high-dimensional real-world optimization problems are becoming realities hence PSO algorithm therefore needs some reworking to enhance it for better performance in handling such problems. This paper proposes a new PSO variant called Adaptive Velocity PSO (AV-PSO), which adaptively adjusts the velocity of particles based on Euclidean distance between the position of each particle and the position of the global best particle. To avoid getting trapped in local optimal, chaotic characteristics was introduced into the particle position updating formula. In all experiments, it is shown that AV-PSO is very efficient for solving low and high-dimensional global optimization problems. Empirical results show that AV-PSO outperformed AIWPSO, PSOrank, CRIW-PSO, def-PSO, e1-PSO and APSO. It also performed better than LSRS in many of the tested high-dimensional problems. AV-PSO was also used to optimize some high-dimensional problems with 4,000 dimensions with very good results.