Scholarlink

<p>Energy-efficient scheduling methods are essential for reducing energy consumption in energy-intensive industries. This study addresses a multi-objective scheduling problem for a single machine in investment casting, aiming to minimize total energy consumption and makespan. It explores the potential of utilizing afterheat from the melting furnace to accelerate the melting process. To solve this multi-objective scheduling problem, we employ a Deep Q-network (DQN) within a Reinforcement Learning (RL) framework. During algorithm implementation, we apply the Taguchi method to reduce the action space, following the “less is more” principle. Additionally, we use the Random Forest approach to predict reward parameters, addressing the challenge of uncertain reward determination. To validate the effectiveness of the proposed method, we generate 54 test instances and establish three evaluation metrics for comparative experiments against heuristic and meta-heuristic methods. The results show that the proposed method achieves win rates of 87%, 44%, and 100% in the T-ratio, E-ratio, and W-T-E-ratio metrics compared to the heuristic method. Additionally, it outperforms the meta-heuristic method in most instances in W-T-E-ratio metric.</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

A deep reinforcement learning approach to energy-efficient scheduling with afterheat recovery

  • 期刊论文

  • 发表日期：

    2025年5月30日

查看全文
  • Conghui Wang,
  • Ge Yang,
  • Kim Mee Chong

摘要

Energy-efficient scheduling methods are essential for reducing energy consumption in energy-intensive industries. This study addresses a multi-objective scheduling problem for a single machine in investment casting, aiming to minimize total energy consumption and makespan. It explores the potential of utilizing afterheat from the melting furnace to accelerate the melting process. To solve this multi-objective scheduling problem, we employ a Deep Q-network (DQN) within a Reinforcement Learning (RL) framework. During algorithm implementation, we apply the Taguchi method to reduce the action space, following the “less is more” principle. Additionally, we use the Random Forest approach to predict reward parameters, addressing the challenge of uncertain reward determination. To validate the effectiveness of the proposed method, we generate 54 test instances and establish three evaluation metrics for comparative experiments against heuristic and meta-heuristic methods. The results show that the proposed method achieves win rates of 87%, 44%, and 100% in the T-ratio, E-ratio, and W-T-E-ratio metrics compared to the heuristic method. Additionally, it outperforms the meta-heuristic method in most instances in W-T-E-ratio metric.