Towards a multilevel ant colony optimization
Abstract
Ant colony optimization is a metaheuristic approach for solving combinatorial optimization problems which belongs to swarm intelligence techniques. Ant colony optimization algorithms are one of the most successful strands of swarm intelligence which has already shown very good performance in many combinatorial problems and for some real applications. This thesis introduces a new multilevel approach for ant colony optimization to solve the NP-hard problems shortest path and traveling salesman. We have reviewed different elements of multilevel algorithm which helped us in construction of our proposed multilevel ant colony optimization solution. We for comparison purposes implemented our own multi-threaded variant Dijkstra for solving shortest path to compare it with single level and multilevel ant colony optimization and reviewed different techniques such as genetic algorithms and Dijkstra’s algorithm. Our proposed multilevel ant colony optimization was developed based on the single level ant colony optimization which we both implemented. We have applied the novel multilevel ant colony optimization to solve the shortest path and traveling salesman problem. We show that the multilevel variant of ant colony optimization outperforms single level. The experimental results conducted demonstrate the overall performance of multilevel in comparison to the single level ant colony optimization, displaying a vast improvement when employing a multilevel approach in contrast to the classical single level approach. These results gave us a better understanding of the problems and provide indications for further research.
Description
Masteroppgave i Informasjons- og kommunikasjonsteknologi IKT590 Universitetet i Agder 2014