Abstract

This thesis will use the traveling salesman problem (TSP) as a tool to help present and investigate several new techniques that improve the overall performance of genetic algorithms (GA). Improvements include a new parent selection algorithm, harem select, that outperforms all other parent selection algorithms tested, some by up to 600%. Other techniques investigated include population seeding, random restart, heuristic crossovers, and hybrid genetic algorithms, all of which posted improvements in the range of 1% up to 1100%. Also studied will be a new algorithm, GRASP, that is just starting to enjoy a lot of interest in the research community and will also been applied to the traveling salesman problem (TSP). Given very little time to run, relative to other popular metaheuristic algorithms, GRASP was able to come within 5% of optimal on several of the TSPLIB maps used for testing. Both the GA and the GRASP algorithms will be compared with commonly used metaheuristic algorithms such as simulated annealing (SA) and reactive tabu search (RTS) as well as a simple neighborhood search - greedy search.

Library of Congress Subject Headings

Traveling-salesman problem; Genetic algorithms; Heuristic programming

Publication Date

2006

Document Type

Thesis

Department, Program, or Center

Computer Science (GCCIS)

Advisor

Anderson, Peter

Comments

Note: imported from RIT’s Digital Media Library running on DSpace to RIT Scholar Works. Physical copy available through RIT's The Wallace Library at: QA402.6 .S54 2006

Campus

RIT – Main Campus

Download

Share

COinS