MISTA 2009 HyFlex: A Flexible Framework for the Design and Analysis of Hyper-heuristics Edmund K. Burke · Tim Curtois · Matthew Hyde · Graham Kendall · Gabriela Ochoa · Sanja Petrovic · Jos´ e Antonio V´ azquez-Rodr´ ıguez 1 Introduction Despite the success of heuristic search methods in solving real-world computational search problems, it is often still difficult to easily apply them to new problems, or even new instances of similar problems. These difficulties arise mainly from the sig- nificant number of parameter or algorithm choices involved when using these type of approaches, and the lack of guidance as to how to proceed when selecting them. Hyper- heuristics are an emergent search methodology, the goal of which is to automate the process of either (i) selecting and combining simpler heuristics [5], or (ii) generating new heuristics from components of existing heuristics [6]; in order to solve hard com- putational search problems. The main motivation behind hyper-heuristics is to raise the level of generality in which search methodologies can operate. They can be dis- tinguished from other heuristic search algorithms, in that they operate on a search space of heuristics (or heuristic components) rather than directly on the search space of solutions to the underlying problem. The hyper-heuristic framework presented in [5,10], operates at a high level of ab- straction and often has no knowledge of the domain. It only has access to a set of low- level heuristics (neighbourhood structures) that it can call upon, and has no knowledge of the functioning of those low-level heuristics. The motivation behind this approach is that once a hyper-heuristic algorithm has been developed, it can be applied to a new problem by replacing the set of low level heuristics and the evaluation function. Figure 1 illustrates that there is a barrier between the low-level heuristics and the hyper-heuristic. Domain knowledge is not allowed to cross this barrier. Therefore, the hyper-heuristic has no knowledge of the domain under which it is operating. It only knows that it has n low-level heuristics on which to call, and it knows it will be passed the results of a given solution once it has been evaluated by the evaluation function. A well defined interface between the hyper-heuristic layer and the problem domain layer needs to be provided, which will allow both the communication between the high-level strategy and the low-level heuristics, and the interchange of relevant non-domain in- formation between the two layers. Furthermore, such an interface would permit the rapid incorporation of new problem domains. In other words, once a new domain is Automated Scheduling, Optimisation and Planning (ASAP) Group, School of Computer Sci- ence, University of Nottingham, UK E-mail: {ekb,tec,mvh,gxk,gxo,sxp,jav}@cs.nott.ac.uk