Peckish Initialisation Strategies for Evolutionary Timetabling

Some evolutionary algorithm (EA)/timetabling researchers find benefit from combining an EA with graph-colouring based greedy algorithms, while others opt for a simpler but faster method. We consider a combination of the two approaches, largely retaining the speed of the simpler method while adopting the greedy method to bootstrap the process. In this combination, the initial population is produced by a ‘peckish’ timetable construction algorithm, similar to a greedy algorithm, but less concerned with finding a best timeslot for an event at each step. We find peckish population initialisation more effective than either greedy or random initialisation on non-trivial problems. Peckish initialisation is shown to aid a simple hill-climbing approach in a similar way. Finally, we add to the growing observation that hill-climbing often outperforms an EA on timetabling problems, but that this effect is reversed on problems of particular overconstrainedness or difficulty.

[1]  Peter Ross,et al.  Fast Practical Evolutionary Timetabling , 1994, Evolutionary Computing, AISB Workshop.

[2]  Ben Paechter,et al.  Two solutions to the general timetable problem using evolutionary methods , 1994, Proceedings of the First IEEE Conference on Evolutionary Computation. IEEE World Congress on Computational Intelligence.

[3]  Ben Paechter,et al.  An evolutionary approach to the general timetable problem. , 1993 .

[4]  Marco Dorigo,et al.  Genetic Algorithms and Highly Constrained Problems: The Time-Table Case , 1990, PPSN.

[5]  Peter Ross,et al.  Improving Evolutionary Timetabling with Delta Evaluation and Directed Mutation , 1994, PPSN.

[6]  David R. Jefferson,et al.  Selection in Massively Parallel Genetic Algorithms , 1991, ICGA.