A comparison study of genetic codings for the traveling salesman problem

In applying the genetic algorithm (GA) to optimization problems, both a genetic coding method and a method of genetic operations are essential for making a search effective. Moreover, freedom in a genetic representation, e.g. redundant coding, is indispensable for achieving a successful self-organization in GA. This paper treats the case of the application of a GA to the traveling salesman problem (TSP), and proposes four ways of redundantly coding a tour plan. Then, based on several computational experiments, the coding methods have been mutually compared from the viewpoints of the search efficiency, i.e. the effects of genetic operators, the quality of the obtained tours, and the number of generations required for finding near-optimal tours. As a result, the search for the optimal tour is found to be most effective in the case of the coding based on the link information, while the simple GA is found not to be sufficient for solving large-scale problems. Then, the GA is modified by supplementing some new mechanisms. The results of the computational experiments suggest the applicability of the modified GA to large-scale problems