计算机集成制造系统 ›› 2019, Vol. 25 ›› Issue (第11): 2935-2942.DOI: 10.13196/j.cims.2019.11.022

• 当期目次 • 上一篇    下一篇

基于强化学习的生产再决策问题

夏金,孙宏波,孙立民+   

  1. 烟台大学计算机与控制工程学院
  • 出版日期:2019-11-30 发布日期:2019-11-30
  • 基金资助:
    烟台市科技计划资助项目(2016ZH065)。

Reinforcement learning for production reschedule

  • Online:2019-11-30 Published:2019-11-30
  • Supported by:
    Project supported by the Science & Technology Plan of Yantai City,China(No.2016ZH065).

摘要: 为解决制造型企业面临的订单变更后生产再决策问题,提出一种基于强化学习的生产再决策方案。对订单变更问题使用半马尔可夫决策模型建模,综合考虑企业实际生产环节中的设备使用情况、产品的收益、库存开销以及订单的违约赔偿等因素,将企业收益最大化和变更前后整体生产决策差异最小化作为优化目标,采用动态改变探索速率和学习速率的Q-learning算法对生产再决策问题进行优化。数值实验证明,所提方法可以快速解决生产再决策问题。

关键词: 订单变更, 生产再决策, 强化学习, Q-learning算法

Abstract: For the problem of production reschedule of manufacturing industry,a method was proposed based on reinforcement learning.The maximizing manufacturer profit and minimizing changes of existed production plans were set to be optimal objective,so that the equipment conditions,profit,storage cost and default cost of industry in production process could be balanced.The Q-learning algorithm which dynamically had changed the exploration rate and the learning rate was employed to optimize the problem of production reschedule.The numerical experiments showed that the optimal production reschedule plan could be quickly obtained with the proposed method.

Key words: order change, production reschedule, reinforcement learning, Q-learning algorithm

中图分类号: