Scholarlink

Efficient multi-satellite resource scheduling is critical for maximizing constellation utilization, ensuring timely task completion, and optimizing energy allocation in dynamic space environments. In this work, a novel satellite beam resource scheduling algorithm that integrates Graph Attention Networks with Double Deep Q-learning is proposed. The method addresses the dynamic multi-satellite scheduling problem by modeling satellite-beam relationships as a graph structure, where nodes represent satellite-beam pairs and edges capture operational dependencies. The graph attention mechanism adaptively learns task priorities and resource constraints, while temporal convolutional layers extract time-series features from beam status data. Experimental results demonstrate that the proposed algorithm achieves 2.99% improvement in average reward and 2.04% increase in successful task completion compared to baseline DDQTS, with a 98.8% success rate.

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

Graph Attention Double Deep Q-Learning For Dynamic Task Scheduling in Multisatellite Resource Allocation

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Jiayu Qu,
  • Wenjing Wu,
  • Kaixin Cui,
  • Dawei Shi

摘要

Efficient multi-satellite resource scheduling is critical for maximizing constellation utilization, ensuring timely task completion, and optimizing energy allocation in dynamic space environments. In this work, a novel satellite beam resource scheduling algorithm that integrates Graph Attention Networks with Double Deep Q-learning is proposed. The method addresses the dynamic multi-satellite scheduling problem by modeling satellite-beam relationships as a graph structure, where nodes represent satellite-beam pairs and edges capture operational dependencies. The graph attention mechanism adaptively learns task priorities and resource constraints, while temporal convolutional layers extract time-series features from beam status data. Experimental results demonstrate that the proposed algorithm achieves 2.99% improvement in average reward and 2.04% increase in successful task completion compared to baseline DDQTS, with a 98.8% success rate.