A Robust Multi-Agent Reinforcement Learning Approach for Scheduling Problems

Scheduling is a decision making process that is used on a regular basis in many real life situations, for example, manufacturing scheduling takes care of the al- location of limited manufacturing resources over time among parallel and sequential manufacturing activities. The vast majority of the research in scheduling has focused on the development of algorithms assuming complete information and a deterministic environment. However, the real world is not so stable, projects may be subject to un- expected events during execution, which may lead to numerous schedule disruptions, for example, resources can become unavailable, new orders can arrive, operations could take longer than expected, that is why it is so important to develop algorithms able to deal with such disruptions without having to reschedule. In this paper, different types of approaches to deal with uncertainty are studied. Based on this, a new approach to incorporate robustness in the schedule construction process is proposed. Some ex- periments are developed in order to show how the new approach works and how it compares to the existing techniques.

Publication year:2013

Keywords:multi-agents, scheduling

Accessibility:Open