Scholarlink

With the advancement of global carbon neutrality goals, mining transportation, as a critical component generating approximately 50%–60% of operational costs, faces tremendous optimization pressure. Traditional fleet dispatching methods exhibit fundamental limitations in large-scale applications, including insufficient temporal information processing capabilities, transportation load imbalance, and excessive complexity in real-time scheduling decisions. This study proposes a hybrid optimization method based on LSTM-PPO to address these challenges. The method features a framework combining temporal memory modeling with strategy optimization, utilizing LSTM networks to extract temporal features of vehicle states for selective memory of historical dispatching experience, coupled with PPO algorithms for stable policy learning to meet performance and responsiveness requirements. Experiments are conducted based on simulation environments constructed from real mining scenarios, with performance validation using a fleet size of 20 vehicles. Experimental results demonstrate that compared to real operational data, the proposed method achieves a 3.0% reduction in cost control while improving transportation volume by 25.0%, reducing queue time by 21.0%, and enhancing system stability by 144.0%. The method is effective and efficient in ensuring the performance and responsiveness of large-scale heterogeneous fleet dispatching, providing a solution that balances efficiency and equilibrium for intelligent fleet optimization in complex dynamic environments.

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

Large-scale Mining Fleet Dispatching Optimization Based on LSTM-PPO

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Haonan Zhang,
  • Jue Yang,
  • Yong Zhang,
  • Mingan Zhao,
  • Yiting Kang,
  • Yanbiao Feng,
  • Xinxin Zhao,
  • Xuan Zhang

摘要

With the advancement of global carbon neutrality goals, mining transportation, as a critical component generating approximately 50%–60% of operational costs, faces tremendous optimization pressure. Traditional fleet dispatching methods exhibit fundamental limitations in large-scale applications, including insufficient temporal information processing capabilities, transportation load imbalance, and excessive complexity in real-time scheduling decisions. This study proposes a hybrid optimization method based on LSTM-PPO to address these challenges. The method features a framework combining temporal memory modeling with strategy optimization, utilizing LSTM networks to extract temporal features of vehicle states for selective memory of historical dispatching experience, coupled with PPO algorithms for stable policy learning to meet performance and responsiveness requirements. Experiments are conducted based on simulation environments constructed from real mining scenarios, with performance validation using a fleet size of 20 vehicles. Experimental results demonstrate that compared to real operational data, the proposed method achieves a 3.0% reduction in cost control while improving transportation volume by 25.0%, reducing queue time by 21.0%, and enhancing system stability by 144.0%. The method is effective and efficient in ensuring the performance and responsiveness of large-scale heterogeneous fleet dispatching, providing a solution that balances efficiency and equilibrium for intelligent fleet optimization in complex dynamic environments.