Design of vehicle routing problem domains for a hyper-heuristic framework

The branch of algorithms that uses adaptive methods to select or tune heuristics, known as hyper-heuristics, is one that has seen a large amount of interest and development in recent years. With an aim to develop techniques that can deliver results on multiple problem domains and multiple instances, this work is getting ever closer to mirroring the complex situations that arise in the corporate world. However, the capability of a hyper-heuristic is closely tied to the representation of the problem it is trying to solve and the tools that are available to do so. This thesis considers the design of such problem domains for hyper-heuristics. In particular, this work proposes that through the provision of high-quality data and tools to a hyper-heuristic, improved results can be achieved. A definition is given which describes the components of a problem domain for hyper-heuristics. Building on this definition, a domain for the Vehicle Routing Problem with Time Windows is presented. Through this domain, examples are given of how a hyper- heuristic can be provided extra information with which to make intelligent search decisions. One of these pieces of information is a measure of distance between solution which, when used to aid selection of mutation heuristics, is shown to improve results of an Iterative Local Search hyper-heuristic. A further example of the advantages of providing extra information is given in the form of the provision of a set of tools for the Vehicle Routing Problem domain to promote and measure ’fairness’ between routes. By offering these extra features at a domain level, it is shown how a hyper-heuristic can drive toward a fairer solution while maintaining a high level of performance.