Scholarlink

<p>The application of automated guided vehicles (AGVs) is essential for improving production efficiency and flexibility in modern manufacturing systems. However, task scheduling in assembly workshops using AGVs presents significant challenges, particularly when AGV resources are limited. To address this issue, this study proposes a dynamic scheduling framework using deep reinforcement learning (DRL). The scheduling problem is modeled mathematically, with a decision-making system designed using tailored reward functions, state spaces, and action spaces. The framework integrates the deep Q-network with noisy networks (noisy_DQN), utilizing noise-enhanced exploration to enhance learning efficiency and decision accuracy. Experimental results demonstrate noisy_DQN significantly outperforms five commonly used DRL algorithms in stability, convergence, and optimization performance. Specifically, the average relative percentage deviation (ARPD) of noisy_DQN is over three times lower than that of the other algorithms. By refining reward functions and introducing a hierarchical state space design, the noisy_DQN framework enables precise scheduling. This method significantly improves AGV utilization, reduces task delays, and offers a robust solution to scheduling problems in assembly workshops.</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

Dynamic scheduling for an assembly workshop with insufficient AGVs using deep reinforcement learning

  • 期刊论文

  • 发表日期：

    2025年8月2日

查看全文
  • Zhigang Xu,
  • Hailong Song,
  • Shengluo Yang,
  • Shuoxin Yin,
  • Junyi Wang

摘要

The application of automated guided vehicles (AGVs) is essential for improving production efficiency and flexibility in modern manufacturing systems. However, task scheduling in assembly workshops using AGVs presents significant challenges, particularly when AGV resources are limited. To address this issue, this study proposes a dynamic scheduling framework using deep reinforcement learning (DRL). The scheduling problem is modeled mathematically, with a decision-making system designed using tailored reward functions, state spaces, and action spaces. The framework integrates the deep Q-network with noisy networks (noisy_DQN), utilizing noise-enhanced exploration to enhance learning efficiency and decision accuracy. Experimental results demonstrate noisy_DQN significantly outperforms five commonly used DRL algorithms in stability, convergence, and optimization performance. Specifically, the average relative percentage deviation (ARPD) of noisy_DQN is over three times lower than that of the other algorithms. By refining reward functions and introducing a hierarchical state space design, the noisy_DQN framework enables precise scheduling. This method significantly improves AGV utilization, reduces task delays, and offers a robust solution to scheduling problems in assembly workshops.