计算机工程与应用 ›› 2020, Vol. 56 ›› Issue (13): 150-155.DOI: 10.3778/j.issn.1002-8331.1903-0230

• 模式识别与人工智能 • 上一篇    下一篇

基于迁移学习的多任务分配算法

王梦娇,尹翔,黄宁馨   

  1. 扬州大学 信息工程学院,江苏 扬州 225009
  • 出版日期:2020-07-01 发布日期:2020-07-02

Multitask Assignment Algorithm Based on Transfer Learning

WANG Mengjiao, YIN Xiang, HUANG Ningxin   

  1. College of Information Engineering, Yangzhou University, Yangzhou, Jiangsu 225009, China
  • Online:2020-07-01 Published:2020-07-02

摘要:

对于多任务分配问题,传统的方法针对每一个任务独立地寻找一个最优分配方案,没有考虑任务间的关联以及历史经验对新任务分配的影响,因而复杂度较高。研究了多智能体系统中的多任务分配问题,通过迁移学习来加速任务分配以及子任务的完成。在分配目标任务时,通过计算当前任务和历史任务的相似度找到最适合的源任务,再将源任务的分配模式迁移到目标任务中,并在完成子任务的过程中使用迁移学习,从而提高效率,节约时间。最后,通过“格子世界”的实验证明了该算法在运行时间和平均带折扣回报方面都优于基于Q学习的任务分配算法。

关键词: 多智能体系统, 任务分配, 迁移学习, Q学习

Abstract:

To multitask assignment problem, traditional methods search for an optimal solution without considering the relationship among tasks as well as the impact of historical experience to new tasks, which leads to high complexity. This paper studies the issue of multitask allocation in multiagent systems. Transfer learning is employed to accelerate task allocation and execution. When assigning the target task, the most suitable source task is found by calculating the similarity between source tasks and target task. Afterwards, the allocation mode and execution process of source task are transferred to the new task to improve efficiency and save time. Finally, the empirical results reveal that the proposed approach outperforms the existing methods in terms of running time and the average discounted return.

Key words: multi-agent system, task allocation, transfer learning, Q-learning