Reinforcement learning with EGD based hyper heuristic system for exam timetabling problem

Scheduling problems such as nurse rostering problems, university timetabling, arise in almost all areas of human activity. As a result, there are many methods to solve them. Some of the most effective techniques on the benchmark data are Meta heuristic methods. Unfortunately, these methods rely upon either parameter tuning or deep understanding of domain knowledge. They are not capable of dealing with other different problems. Thus, this has led to the development of hyper heuristics system. One contribution of this paper is to attempt to use the extended great deluge (EGD) method as a move acceptance method to drive the selection of low level heuristic within hyper heuristic (HH) framework. Moreover, hyper heuristic search with memory, which is also used to store the accepted solutions at each iteration, is also applied. The proposed EGD based HH is tested to a benchmark set of examination timetabling problem as an instance of a constraint based real world optimization problem and the experiment results are also shown.