Parallel Solution of Large Scale Traveling Salesman Problems by using Clustering and Evolutionary Algorithms

Abstract

Evolutionary Algorithms provide reasonable speed in time for solving optimization problems. However, these approaches may stay insufficient when the problem gets bigger and needs more hardware resource. Because it is not feasible to improve memory and computation power depending on the problem size, there is a need for developing new techniques in order to solve these optimization problems in less time with smaller error ratios. In this study, it is aimed to solve the Traveling Salesman Problem, one of the NP-complete complexity problems, by partitioning the problem with a clustering technique, K-Means, and solving these pieces with Genetic Algorithm and finally combining these solutions into one. As the experimental results suggest, in comparison to solving large scale optimization problems as single problems, solving them by partitioning them yields more convincing results in both solution quality and time. In addition, it is observed that the performance of the technique yields better as the problem size gets bigger.