A GP Hyper-Heuristic Approach for Generating TSP Heuristics

Abstract

A wide range of heuristics has been developed over the last decades as a way to obtain good quality solutions in reasonable time on large scale optimisation problems. However, heuristics are problem specific, i.e. lack of generalisation potential, while requiring time to design. Hyper-heuristics have been proposed to address these limitations by directly searching in the heuristics' space. This work more precisely focuses on a heuristic generation method, as opposed to heuristic selection, for the travelling salesman problem (TSP). Learning is achieved with a genetic programming (GP) approach, for which novel specific terminals are introduced. The performance of the proposed GP hyper-heuristic is evaluated on a large set of TSP instances and compared to state-of-the-art heuristics. Experiments demonstrate that the generated heuristics are outperforming existing ones while having similar or lower complexity.