HySST: Hyper-heuristic Search Strategies and Timetabling

High school timetabling (HST) is a well-known real-world combinatorial optimisation problem. It requires the scheduling of events and resources, such as courses, classes of students, teachers, rooms and more within a fixed number of time slots subject to a set of constraints. In a standard fashion, constraints are separated into ‘hard’ and soft. The hard constraints must be satisfied in order to achieve feasibility, whereas the soft constraints represent preferences and a solution for a given problem; solutions are expected to satisfy all hard constraints and as many soft constraints as possible. The HST problem is known to be NP-complete [2] even in simplified forms. For a recent survey of HST see [4]. Also, see [5] for a description of the specific HST version studied here, and also of the third international timetabling competition, ITC2011. Briefly, the ITC2011 problem instances contain 15 types of constraints and a candidate solution is evaluated in terms of two components: feasibility and preferences. The evaluation function computes the weighted hard and soft constraint violations for a given solution as infeasibility and objective values, respectively. For the comparison of algorithms, a solution is considered to be better than another if it has a smaller infeasibility value, or an equal infeasibility value and a smaller objective value. Our approach to solving the HST is based on development of a hyperheuristic (see [1] for a recent survey) in order to intelligently control the application of a range of neighbourhood move operators. Hyper-heuristics explore the space of (meta-)heuristics, as opposed to directly searching the space of solutions, and generally, split into one of two types: selection hyperheuristics that select between existing low-level heuristics, and generation