Investigation of Initialization Methods for Particle Swarm Optimization of Finite-thrust Orbital Transfers
Open Access
Author:
Honeychuck, Matthew C
Area of Honors:
Aerospace Engineering
Degree:
Bachelor of Science
Document Type:
Thesis
Thesis Supervisors:
Robert Graham Melton, Thesis Supervisor Robert Graham Melton, Thesis Honors Advisor David Bradley Spencer, Faculty Reader Dr. George A Lesieutre, Faculty Reader
Keywords:
particle swarm optimization finite thrust orbital transfer Sobol
Abstract:
Particle Swarm Optimization is an evolutionary numerical optimization method that has proven capable of finding global optimum solutions when applied to a wide variety of problems. The algorithm takes advantage of information sharing between particles in a population, or swarm, to converge to a solution. In this thesis, the algorithm is applied to the problem of a finite thrust orbital transfer between two co-planar circular orbits, and is solved for three different ratios of orbital radii. Different methods of initializing the swarm are investigated to determine any performance benefits. Three different techniques for creating the initial swarm are analyzed: uniform random number generation, and two variations of quasi-random number generation using Sobol sequences. Additionally, initial swarm sizes of 50, 100, 200, 500, 1000, 2000 and 3000 particles are used. After running all cases, results generally suggest that increasing the initial swarm size improves the performance of the PSO algorithm. The best method of swarm creation appears to be problem dependent. Future work should consider a larger number of cases to better define trends in the data and, if possible, reveal more general conclusions.