摘要
近来实时动态任务分配机制得到越来越多的研究.考虑多任务流并存时的任务分配问题,提出基于Q学习的分布式多任务流调度算法,不仅能适应自身任务流的到达过程,还充分兼顾其他任务流的到达及分配的影响,从而使得整个系统长期期望回报最大.分布式特性使得算法适用于开放的,局部可见的多Agent系统;强化学习的采用使得任务分配决策自适应系统环境隐藏的不确定性.实验表明此算法具有较高的任务吞吐量和任务完成效率.
Recently real-time dynamic task allocation mechanisms draw more attention.Allocation of multiple task flows are considered in this paper,and a distributed scheduling algorithm based on Q-learning for multiple task flows is proposed.This algorithm can not only adapt to task flow on itself,but also take arrival and allocation of other task flows into account,thereby maximizing long-term expected reward of the whole system.Distributed property renders it applicable to open multi-agent systems with local visibility,while reinforcement learning makes allocation decisions adaptable to environment uncertainty.Experiments establish that the algorithm has higher task throughput,improving system efficiency.
出处
《小型微型计算机系统》
CSCD
北大核心
2010年第4期597-602,共6页
Journal of Chinese Computer Systems
关键词
Agent合作
任务分配
多任务流
Q学习
agent cooperation
task allocation
multiple task flows
Q-learning