pythOPT: A problem-solving environment for optimization methods
Date
2017-02-15
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
ORCID
0000-0002-2961-1193
Type
Thesis
Degree Level
Masters
Abstract
Optimization is a process of finding the best solutions to problems based on mathematical
models. There are numerous methods for solving optimization problems,
and there is no method that is superior for all problems. This study focuses on the
Particle Swarm Optimization (PSO) family of methods, which is based on the swarm
behaviour of biological organisms. These methods are easily adjustable, scalable, and
have been proven successful in solving optimization problems.
This study examines the performance of nine optimization methods on four sets
of problems. The performance analysis of these methods is based on two performance
metrics (the win-draw-loss metric and the performance profiles metric) that are used
to analyze experimental data. The data are gathered by using each optimization
method in multiple configurations to solve four classes of problems.
A software package pythOPT was created. It is a problem-solving environment
that is comprised of a library, a framework, and a system for benchmarking optimization
methods. pythOPT includes code that prepares experiments, executes
computations on a distributed system, stores results in a database, analyzes those
results, and visualizes analyses. It also includes a framework for building PSO-based
methods and a library of benchmark functions used in one of the presented analyses.
Using pythOPT, the performance of these nine methods is compared in relation
to three parameters: number of available function evaluations, accuracy of solutions,
and communication topology. This experiment demonstrates that two methods
(SPSO and GCPSO) are superior in finding solutions for the tested classes of
problems. Finally, by using pythOPT we can recreate this study and produce similar
ones by changing the parameters of an experiment. We can add new methods and
evaluate their performances, and this helps in developing new optimization methods.
Description
Keywords
optimization, pso, particle swarm optimization, direct, benchmarking
Citation
Degree
Master of Science (M.Sc.)
Department
Computer Science
Program
Computer Science