基于Q-学习的动态单机调度

Q-Learning Based Dynamic Single Machine Scheduling

针对当前基于Q-学习的Agent生产调度优化研究甚少的现状,利用Q-学习对动态单机调度问题在3种不同系统目标下的调度规则动态选择问题进行了研究.在建立Q-学习与动态单机调度问题映射机制的基础上,通过MATLAB实验仿真,对算法性能进行了评价.仿真结果表明,对于不同的系统调度目标,Q-学习能提高Agent的适应能力,达到单一调度规则无法达到的性能,适合基于Agent的动态生产调度环境.

Q-learning was applied to a dynamic single-machine scheduling problem. Corresponding to the environment status change and three predefined system performance measurement, the machine agent that is embedded with Q-learning can select an appropriate dispatching rule dynamically. Based on the model between Q-learning and the dynamic single-machine scheduling problem, the performance of Q-learning was evaluated through simulations in MATLABa environment. The simulation results demonstrate that Q-learning can perform well for different system objectives, which is impossible for single dispatching rule. Therefore, Q-learning is promising for application to the agent-based dynamic production scheduling.