Scholarlink

<p>The flexible job-shop scheduling problem (FJSP) is a critical domain within production planning and control. It has garnered significant interest due to its capacity to optimize resource allocation, enhance production efficiency, reduce manufacturing costs, and bolster the coordination of the entire production process. As manufacturing systems evolve to incorporate advanced technologies, the integration of automated guided vehicles into these systems introduces a new layer of complexity to scheduling. The FJSP with automated guided vehicles (FJSP-AGV) considers the transportation time of AGVs and adds the complexity of AGV allocation to the FJSP, significantly increasing the uncertainty and complexity of the scheduling process. This paper proposes a scheduling optimization algorithm based on a heterogeneous graph neural network that integrates multiple attention mechanism, designed within an end-to-end learning framework to achieve scheduling of FJSP-AGV. First, a heterogeneous graph model of the workshop state is construct to extract the complex relationships among operations, machines, and AGVs. Next, we utilize multiple attention mechanism to process the state features. Subsequently, we optimize the decision model's performance using the proximal policy optimization (PPO) algorithm. Finally, the algorithm framework is trained using randomly generated cases, and the trained model is combined with Monte Carlo Tree Search (MCTS) and applied to both standard cases and randomly generated cases. The results demonstrate that, in standard cases, our algorithm outperforms existing mainstream reinforcement learning algorithms, while also exhibiting superior generalization in randomly generated cases.</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

Deep reinforcement learning-driven graph attention network for flexible job-shop scheduling with automated guided vehicles

  • 期刊论文

  • 发表日期：

    2026年4月24日

查看全文
  • Minghai Yuan,
  • Qi Yu,
  • Songwei Lu,
  • Fengque Pei,
  • Xiaoting Wang

摘要

The flexible job-shop scheduling problem (FJSP) is a critical domain within production planning and control. It has garnered significant interest due to its capacity to optimize resource allocation, enhance production efficiency, reduce manufacturing costs, and bolster the coordination of the entire production process. As manufacturing systems evolve to incorporate advanced technologies, the integration of automated guided vehicles into these systems introduces a new layer of complexity to scheduling. The FJSP with automated guided vehicles (FJSP-AGV) considers the transportation time of AGVs and adds the complexity of AGV allocation to the FJSP, significantly increasing the uncertainty and complexity of the scheduling process. This paper proposes a scheduling optimization algorithm based on a heterogeneous graph neural network that integrates multiple attention mechanism, designed within an end-to-end learning framework to achieve scheduling of FJSP-AGV. First, a heterogeneous graph model of the workshop state is construct to extract the complex relationships among operations, machines, and AGVs. Next, we utilize multiple attention mechanism to process the state features. Subsequently, we optimize the decision model's performance using the proximal policy optimization (PPO) algorithm. Finally, the algorithm framework is trained using randomly generated cases, and the trained model is combined with Monte Carlo Tree Search (MCTS) and applied to both standard cases and randomly generated cases. The results demonstrate that, in standard cases, our algorithm outperforms existing mainstream reinforcement learning algorithms, while also exhibiting superior generalization in randomly generated cases.