现代制造工程 ›› 2024, Vol. 530 ›› Issue (11): 26-36.doi: 10.16731/j.cnki.1671-3133.2024.11.004

• 先进制造系统管理运作 • 上一篇    下一篇

IRF-RL的混合流水车间动态调度方法研究*

张梦杰1, 杨晓英2,3, 李博2   

  1. 1 河南科技大学商学院,洛阳 471000;
    2 河南科技大学机电工程学院,洛阳 471003;
    3 机械装备先进制造河南省协同创新中心,洛阳 471003
  • 收稿日期:2024-01-02 出版日期:2024-11-18 发布日期:2024-11-29
  • 作者简介:张梦杰,硕士研究生,主要研究方向为智能算法与车间调度。E-mail:qwaszx42669287@163.com;杨晓英,博士研究生导师,主要研究方向为工业工程与智能制造等。E-mail:lyyxy@haust.edu.cn;李博,硕士研究生,主要研究方向为智能优化算法。E-mail:18730316960@163.com
  • 基金资助:
    *国家重点研发计划项目(2018YFB1701205);企业委托项目(HX20221116)

Hybrid flow shop dynamic scheduling method based on improved random forest and reinforcement learning

ZHANG Mengjie1, YANG Xiaoying2,3, LI Bo2   

  1. 1 Business School,Henan University of Science and Technology, Luoyang 471000,China;
    2 School of Mechatronics Engineering,Henan University of Science and Technology, Luoyang 471003,China;
    3 Henan Provincial Collaborative Innovation Center for Advanced Manufacturing of Machinery and Equipment, Luoyang 471003,China
  • Received:2024-01-02 Online:2024-11-18 Published:2024-11-29

摘要: 为适应混合流水车间生产需求,提出了一种基于机器学习的两阶段动态调度方法。在离线挖掘阶段,以历史数据为基础,采用改进随机森林算法建立一个由制造系统生产状态到最优调度规则的知识映射网络,挖掘出有价值的调度规则用于在线决策,跳过预热阶段提高调度效率进而优化调度方案;在线调度阶段,采用强化学习算法对车间状态的实时数据进行分析和训练,根据系统状态的动态变化优化策略选择,以实现对扰动事件的自适应和快速响应能力;仿真实验结果验证了结合数据挖掘和强化学习的两阶段动态调度方法具有可行性和有效性,可充分利用制造数据并在线调度制造执行过程。

关键词: 混合流水车间, 动态调度, 强化学习, 改进随机森林, 数据驱动

Abstract: In order to adapt the production demands of hybrid flow shop, a two-stage dynamic scheduling method based on machine learning is proposed.In the offline mining phase, based on historical data, the improved random forest algorithm is used to establish a knowledge mapping network from the production state of the manufacturing system to the optimal scheduling rules, which can be mined for online decision-making, thus skipping the warm-up phase to improve the scheduling efficiency and optimize the scheduling scheme. In the online scheduling phase, the reinforcement learning algorithm is used to analyze and train the real-time data of the flow shop state. And it optimizes the strategy selection according to the system state′s dynamic changes of the system state to achieve the adaptive and fast response capability to the perturbation events. The simulation experiment verified that the two-phase dynamic scheduling approach combining data mining and reinforcement learning is more feasible and effective for fully utilizing manufacturing data and scheduling the manufacturing execution process online.

Key words: hybrid flow shop, dynamic scheduling, reinforcement learning, improved random forest, data-driven

中图分类号: 

版权所有 © 《现代制造工程》编辑部 
地址:北京市东城区东四块玉南街28号 邮编:100061 电话:010-67126028 电子信箱:2645173083@qq.com
本系统由北京玛格泰克科技发展有限公司设计开发 技术支持:support@magtech.com.cn