Analysing the High Level Heuristics within a Graph Based Hyper-heuristic

This paper presents our work on analysing high level heuristics within a graph based hyper-heuristic framework. Local search based methods (Tabu Search and steepest descent method) are compared as the high level heuristics within the graph based hyper-heuristic framework and the analysis of their neighborhood structures and performance on the high level landscape is carried out. An iterated steepest descent based hyper-heuristic is developed based on the analysis undertaken. Experimental results on benchmark exam timetabling problems demonstrate the simplicity and efficiency of this improved hyper-heuristic approach. 1 Exam Timetabling Problem Timetabling problems arise in many real-world problem solving circumstances including educational timetabling (e.g. [23], [24]), nurse rostering (e.g. [2], [38]), sports timetabling (e.g. [30]) and transportation scheduling (e.g. [34]), etc. They have attracted a significant attention in the operational research and artificial intelligence research community during the last 4 decades (e.g. [8], [9], [10], [20], [21]). This paper concentrates upon examination timetabling. Selected reviews of the educational timetabling that have appeared over the years are (e.g. [5], [12], [18], [23], [24], [39], , [40]). An exam timetabling problem consists of assigning a number of exams into a limited number of time periods while satisfying a set of constraints. Real world instances of exam timetabling problems are very complex. Constraints are usually grouped into two types: hard constraints, which need to be satisfied under any circumstances and soft constraints, which are desirable but not essential. Solutions which satisfy the stipulated hard constraint are often called feasible solutions. A wide variety of techniques and algorithms have been studied for exam timetabling. Graph heuristics (e.g. [6], [14]) were studied in the early days of timetabling research and they still appear as a widely employed method particularly when integrated with other approaches. Constraint based approaches have also featured significantly in the exam timetabling literature (e.g. [4], [37]). Much recent research in timetabling concerns meta-heuristic techniques [33] which are very successful over a range of complex timetabling problems. Examples include