Affiliations: Faculty of Computer Science and Management, Wroclaw University of Science and Technology, Poland
Correspondence:
[*]
Corresponding author. Marek Kopel, Faculty of Computer Science and Management, Wroclaw University of Science and Technology, Poland. E-mail: Marek.Kopel@pwr.edu.pl.
Abstract: The paper verifies the usefulness of a parallel and adaptive Ant Colony Communities (ACC) for solving the dynamic Travelling Salesman Problem (DTSP). ACC consists of a set of client colonies with a server to coordinate their work. Each one of the client colonies implements a standard ACO algorithm. The paper contains a detailed analysis of the operation of ACO for static TSP in order to identify its most dominant parameters. Graph Generator is used to modify the distances in TSP. In order to catch up with the constant changes the ACC should work in parallel and to adopt to the current distances. This is accomplished by modifying the number of iterations and changing the size of its internal prospective solutions buffer. Two implementations of ACC are presented: an asynchronous that works on computers connected through a LAN and a synchronous that uses a Hadoop environment. Numerous experiments clearly indicate, that the adaptive, parallel ACC outperforms both standard version of ACO as well as its versions adopted for DTSP. This is especially true for highly dynamic Graph Generators.
Keywords: Dynamic TSP, Ant Colony Community, PACO, immigrant based colonies, ACO
parallel implementation
