Scholarlink

<p>Traditional flexible job shop scheduling problems mainly center on machine flexibility. In real manufacturing systems, workers operate machines with multiple skill levels, resulting in different processing times for jobs. Rationalizing worker scheduling is crucial to improving workshop productivity. Additionally, unexpected dynamic events will disrupt the continuity of production and raise costs. To address these challenges, the focus of the paper is a dynamic flexible job shop scheduling problem considering multi-skilled worker constraints (DFJSPW) with new job insertions. We propose an end-to-end deep reinforcement learning (DRL) approach that learns a combination of scheduling policies, including job sequencing, machine selection, and worker assignment, which can directly and sequentially output the scheduling results of each scheduling object at each decision point. States covering all scheduling resources are extracted, and a graph neural network (GNN) is utilized to encode the graph features of job processing. Furthermore, we construct a three-stage decision model, updating the network parameters via proximal policy optimization (PPO). Experiments are conducted on 27 diverse scenarios with randomly generated instances. The outcomes reveal that our approach notably surpasses existing scheduling methods in both solution runtime and quality. Moreover, the trained decision model maintains superior solution performance and strong generalization on the modified benchmark and dynamic instances.</p>

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

An end-to-end deep reinforcement learning method for dynamic flexible job shop scheduling problem considering multi-skilled worker constraints

  • 期刊论文

  • 发表日期：

    2026年3月6日

查看全文
  • Zenghui Yi,
  • Kaidan Deng,
  • Yong Lei,
  • Jingxing Zhang,
  • Liuran Lu,
  • Qianwang Deng

摘要

Traditional flexible job shop scheduling problems mainly center on machine flexibility. In real manufacturing systems, workers operate machines with multiple skill levels, resulting in different processing times for jobs. Rationalizing worker scheduling is crucial to improving workshop productivity. Additionally, unexpected dynamic events will disrupt the continuity of production and raise costs. To address these challenges, the focus of the paper is a dynamic flexible job shop scheduling problem considering multi-skilled worker constraints (DFJSPW) with new job insertions. We propose an end-to-end deep reinforcement learning (DRL) approach that learns a combination of scheduling policies, including job sequencing, machine selection, and worker assignment, which can directly and sequentially output the scheduling results of each scheduling object at each decision point. States covering all scheduling resources are extracted, and a graph neural network (GNN) is utilized to encode the graph features of job processing. Furthermore, we construct a three-stage decision model, updating the network parameters via proximal policy optimization (PPO). Experiments are conducted on 27 diverse scenarios with randomly generated instances. The outcomes reveal that our approach notably surpasses existing scheduling methods in both solution runtime and quality. Moreover, the trained decision model maintains superior solution performance and strong generalization on the modified benchmark and dynamic instances.