Scholarlink

Dynamic task scheduling is crucial for efficient allocation, but existing approaches face challenges with dynamic environments. Traditional methods cannot effectively handle flexible sequences, while reinforcement learning(RL) often encounters training difficulties due to suboptimal hyperparameter selection and reward design. To address these, this paper proposes Multi-Q3IM, a novel dynamic scheduling algorithm based on multi-agent reinforcement learning (MARL). Based on the QMIX framework, Multi-Q3IM introduces an innovative Match strategy for subtask allocation, integrates a resource-constrained Mask mechanism, and incorporates long-term experience to formulate a reward function minimizing makespan. Experiments show that, compared to conventional methods and existing MARL algorithms, Multi-Q3IM exhibits superior adaptability to dynamic variations, achieving significant makespan reduction.

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

Multi-Q3IM: Job Dynamic Task Scheduling Based on Multi-agent Deep Reinforcement Learning in Resource-Constrained Environments

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Linwei Yao,
  • Kuan Li,
  • Huiying Xu,
  • Xinzhong Zhu,
  • Hongbo Li

摘要

Dynamic task scheduling is crucial for efficient allocation, but existing approaches face challenges with dynamic environments. Traditional methods cannot effectively handle flexible sequences, while reinforcement learning(RL) often encounters training difficulties due to suboptimal hyperparameter selection and reward design. To address these, this paper proposes Multi-Q3IM, a novel dynamic scheduling algorithm based on multi-agent reinforcement learning (MARL). Based on the QMIX framework, Multi-Q3IM introduces an innovative Match strategy for subtask allocation, integrates a resource-constrained Mask mechanism, and incorporates long-term experience to formulate a reward function minimizing makespan. Experiments show that, compared to conventional methods and existing MARL algorithms, Multi-Q3IM exhibits superior adaptability to dynamic variations, achieving significant makespan reduction.