Optimizing Re-scheduling with Fast-Adapting Networked Multi-agent Reinforcement Learning

Re-scheduling is the task to modify the existing schedule to make a new schedule that maximizes an objective function. In this paper, we propose a method to solve the re-scheduling task with a multi-agent reinforcement learning framework. We formalize the re-scheduling task as a multi-agent MDP and apply a networked multi-agent reinforcement learning framework. Furthermore, we add adaptation learning steps to allow the agents to learn quickly after the disruption event occurs. We demonstrate that our proposed method significantly improves the performance compared to the naïve MARL implementation and achieves close-to-optimal performance.