Scholarlink

In the field of Flexible Job Shop Scheduling Problem (FJSP), traditional algorithms struggle with resource allocation and efficiency. We propose a Trajectory-based Reinforcement Learning (TBRL) method using self-attention to serialize processing tasks, enhancing long-term dependency management. By mapping job shop states into a graph structure, TBRL addresses job dependencies, improving task allocation and scheduling. Experimental results demonstrate that, compared to existing algorithms, the TBRL scheduling plan substantially reduces the average completion time and decreases the complexity of computing the optimal scheduling plan. Additionally, the TBRL scheduling plan exhibits exceptional stability, outperforming current algorithms on both BRdata and Hurink datasets.

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

TBRL: Trajectory-Based Reinforcement Learning for Flexible Job-Shop Scheduling Problem

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Zheng Chen,
  • Ruijin Wang,
  • Ye Zhu,
  • Chao Tang,
  • Donglin He,
  • Jiachen Wang,
  • Fengli Zhang

摘要

In the field of Flexible Job Shop Scheduling Problem (FJSP), traditional algorithms struggle with resource allocation and efficiency. We propose a Trajectory-based Reinforcement Learning (TBRL) method using self-attention to serialize processing tasks, enhancing long-term dependency management. By mapping job shop states into a graph structure, TBRL addresses job dependencies, improving task allocation and scheduling. Experimental results demonstrate that, compared to existing algorithms, the TBRL scheduling plan substantially reduces the average completion time and decreases the complexity of computing the optimal scheduling plan. Additionally, the TBRL scheduling plan exhibits exceptional stability, outperforming current algorithms on both BRdata and Hurink datasets.