A Statistical Analysis of Particle Swarm Optimization With and Without Digital Pheromones
Vijay Kalivarapu, Eliot H. Winer · 48th AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics, and Materials Conference · 2007
Particle Swarm Optimization (PSO) is a population based heuristic search method for finding global optimal values in multi-disciplinary design optimization problems. PSO is based on simple social behavior exhibited by birds and insects. Due to its simplicity in implementation, PSO has been increasingly gaining popularity in the optimization community. Previous work by the authors demonstrated superior design space search capabilities of particle swarm through implementing digital pheromones in a regular PSO. Although preliminary results showed substantial performance gains, a quantitative assessment has not yet been made to prove the claim. Through a formal statistical hypothesis testing, this paper attempts to evaluate the performance characteristics of PSO with digital pheromones. Specifically, the authors’ claim that the use of digital pheromones improves the solution quality and solution times are tested using various multi-dimensional unconstrained optimization test problems. Conclusions are drawn based on the results from statistical analysis of these test problems and presented in the paper.