Breeding Perturbed City Coordinates and Fooling Travelling Salesman Heuristic Algorithms

Standard heuristic algorithms for the geometric travelling salesman problem (GTSP) frequently produce poor solutions in excess of 25% above the true optimum. In this paper we present some preliminary work that demonstrates the potential of genetic algorithms (GAs) to perturb city coordinates in such a way that the heuristic is ‘fooled’ into producing much better solutions to the GTSP. Initial results for our GA show that by using the nearest neighbour tour construction heuristic on perturbed coordinate sets it is possible to consistently obtain solutions to within a fraction of a percent of the optimum for problems of several hundred cities.