Scholarlink

<p>Mine ventilation systems face significant challenges in dynamic control due to complex network topologies and uncertain underground environments. This paper proposes an intelligent decision-making framework that synergistically integrates graph neural networks (GNN) with deep reinforcement learning (DRL) for optimal ventilation control. A multi-level hierarchical graph representation method is developed to capture topological structures and spatial dependencies of ventilation networks, while an improved Actor-Critic algorithm with prioritized experience replay enables adaptive policy learning under safety constraints. The GNN encoder extracts graph-structured features that enhance the DRL agent’s state representation, facilitating efficient exploration and decision optimization. Experimental validation on simulation platforms and six-month field deployment in an operational coal mine demonstrate substantial improvements: 34.7% higher cumulative rewards compared to conventional methods, 23.7% reduction in energy consumption, and 98.4% safety compliance rate across diverse operational scenarios. The proposed framework advances intelligent mine ventilation management by simultaneously achieving enhanced safety assurance, improved energy efficiency, and robust adaptability to complex dynamic conditions.</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

Intelligent decision-making for mine ventilation systems based on graph neural network and deep reinforcement learning fusion

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年1月30日

查看全文
  • Kai Zhang,
  • Xijun Yang,
  • Hui Li

摘要

Mine ventilation systems face significant challenges in dynamic control due to complex network topologies and uncertain underground environments. This paper proposes an intelligent decision-making framework that synergistically integrates graph neural networks (GNN) with deep reinforcement learning (DRL) for optimal ventilation control. A multi-level hierarchical graph representation method is developed to capture topological structures and spatial dependencies of ventilation networks, while an improved Actor-Critic algorithm with prioritized experience replay enables adaptive policy learning under safety constraints. The GNN encoder extracts graph-structured features that enhance the DRL agent’s state representation, facilitating efficient exploration and decision optimization. Experimental validation on simulation platforms and six-month field deployment in an operational coal mine demonstrate substantial improvements: 34.7% higher cumulative rewards compared to conventional methods, 23.7% reduction in energy consumption, and 98.4% safety compliance rate across diverse operational scenarios. The proposed framework advances intelligent mine ventilation management by simultaneously achieving enhanced safety assurance, improved energy efficiency, and robust adaptability to complex dynamic conditions.