Proceedings of the 13th International Conference on Project Management and Scheduling.
International Conference on Project Management and Scheduling edition:13 location:Leuven date:1-4 April 2012
The present paper demonstrates how decentralized multi-project scheduling problems can be solved
efficiently by project manager agents playing a simple sequence learning game.
The goal is to minimize the average project delay objective.
Agents learn their activity list locally by using reinforcement learning. Meanwhile, they learn to find a suitable place in the overall sequence of all activity lists. All the projects need to choose a unique place in this sequence, while a mediator agent manages a simple dispersion game.
It is shown that the sequence learning game approach is scalable and that it has a large effect on the average project delay and improves the best known results for all the MPSPLIB instances with about 25%.