Dynamic single machine scheduling using Q-learning agent

Single machine scheduling methods have attracted a lot of attentions in recent years. Most dynamic single machine scheduling problems in practice have been addressed using dispatching rules. However, no single dispatching rule has been found to perform well for all important criteria, and no rule takes into account the status or the other resources of system´s environment. In this research, an intelligent agent-based single machine scheduling system is proposed, where the agent is trained by a new improved Q-learning algorithm. In such scheduling system, agent selects one of appropriate dispatching rules for machine based on available information. The agent was trained by a new simulated annealing-based Q-learning algorithm. The simulation results show that the simulated annealing-based Q-learning agent is able to learn to select the best dispatching rule for different system objectives. The results also indicate that simulated annealing-based Q-learning agent could perform well for all criteria, which is impossible when using only one dispatching rule independently.