A Simulated Annealing Hyper-heuristic for University Course Timetabling

The university course timetabling problem involves assigning a given number of events (including lectures, seminars, labs, tutorials, etc) into a limited number of timeslots and rooms subject to given set of constraints. Two primary hard constraints are that no student should be assigned two events in one timeslot and that capacity and features of rooms should satisfy the requirement of the event. Other constraints can be different from one university to another. For example, some universities might want the timetable to be constructed so that there is a good separation between the courses that a student attends, while other universities may prefer to have consecutive courses. Timetabling is a well-known difficult combinatorial problem. Several techniques have been used to automatically generate university timetabling problems, including graph colouring heuristics (Burke et al., 2004), tabu search (Costa, 1994; Schaerf, 1996), simulated annealing (Thompson and Dowsland, 1996; Kostuch, 2004), evolutionary algorithms (Burke et al., 1998) and case-based reasoning (Burke et al., 2006a). Hyper-heuristic approaches have recently been applied to timetabling problems (Burke et al., 2003; Burke et al., 2006b). In (Burke et al. 2003), a tabu search based hyper-heuristic was applied to both a nurse rostering problem and a university course timetabling problem to demonstrate the increased level of generality of the method. In their approach, the hyper-heuristic dynamically ranks a set of heuristics according to their performance in the search history. A tabu list was incorporated to prevent the selection of some heuristics at certain points in the search. At each iteration, the hyper-heuristic keeps applying the highest "non-tabu" heuristic to the current solution until the stopping criterion is met. Competitive results have been obtained on both problems when compared with other state-of-the-art techniques. In (Burke et al., 2006b), a case-based reasoning hyper-heuristic system was proposed for the course timetabling problem. The system differs from other case-based reasoning systems in that case-based reasoning was used to predict the best heuristic methods that can be