基于强化学习的模式驱动调度系统研究

目前,还没有一种调度规则能够根据系统环境状态的改变来进行自适应调整.对此,提出一种基于智能体的模式驱动调度系统,由智能体和仿真环境两个主要部分构成.其中,智能体将利用强化学习(Q学习算法)进行训练,以提高其动态选择合适调度规则的能力.仿真结果表明,这种模式驱动调度系统能够很好地根据系统环境状态的改变选择出对应的最优调度规则,且其调度性能优于单一调度规则,适合于系统环境不断变化的动态调度.

Pattern driven scheduling system based on reinforcement learning

At present,there was no dispatching rules which help scheduling system to take self-adaptation with environment changes.To deal with this problem,an Agent-based pattern driven scheduling system was proposed,which consisted of two parts: the Agent which was trained by reinforcement learning(Q-learning algorithm) to improve its ability to select appropriate dispatching rule and the simulation environment.Simulation results suggested that the system select the best dispatching rule according to the change of the status of system's environment,and dispatching performance was better than single dispatching rule.Therefore,pattern driven scheduling system was suitable for dynamic scheduling in always changing system's environment.