Scholarlink

With the increasing complexity and uncertainty in flexible manufacturing systems, traditional rigid manufacturing models have struggled to meet the demands for multi-variety and small-batch production, making dynamic distributed flexible job shop scheduling (DDFJSS) a critical research focus. This study aims to optimize scheduling under scenarios with concurrent disturbances (e.g., machine breakdown and urgent order insertions) by minimizing the total weighted tardiness (TWT) as the primary objective. To achieve rapid response to dynamic disturbances, an Industrial Internet of Things (IIoT) framework for dynamic perception was proposed. In addition, a multi-agent collaborative decision-making (MACDM) framework is designed, where IIoT-driven equipment states and production data streams replace traditional simulation inputs. The MACDM trains adaptive rescheduling strategies using the double deep Q network (DDQN) algorithm, which dynamically processes real-time equipment states and production data collected from IIoT networks to optimize resource allocation. Experimental results demonstrate that the MACDM significantly outperforms DDQN-based dynamic scheduling approaches, validating the effectiveness of MACDM framework in enhancing real-time responsiveness and decision accuracy.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A Multi-agent Deep Reinforcement Learning Framework for Solving Dynamic Distributed Flexible Job Shop Scheduling in Industrial Internet of Things

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Minghuang Fang,
  • Zi-Qi Zhang,
  • Bin Qian,
  • Rong Hu

摘要

With the increasing complexity and uncertainty in flexible manufacturing systems, traditional rigid manufacturing models have struggled to meet the demands for multi-variety and small-batch production, making dynamic distributed flexible job shop scheduling (DDFJSS) a critical research focus. This study aims to optimize scheduling under scenarios with concurrent disturbances (e.g., machine breakdown and urgent order insertions) by minimizing the total weighted tardiness (TWT) as the primary objective. To achieve rapid response to dynamic disturbances, an Industrial Internet of Things (IIoT) framework for dynamic perception was proposed. In addition, a multi-agent collaborative decision-making (MACDM) framework is designed, where IIoT-driven equipment states and production data streams replace traditional simulation inputs. The MACDM trains adaptive rescheduling strategies using the double deep Q network (DDQN) algorithm, which dynamically processes real-time equipment states and production data collected from IIoT networks to optimize resource allocation. Experimental results demonstrate that the MACDM significantly outperforms DDQN-based dynamic scheduling approaches, validating the effectiveness of MACDM framework in enhancing real-time responsiveness and decision accuracy.