最新刊期

      He Xian

      DOI：10.11959/j.issn.2096-8930.tdyt2026039
      摘要：To address the scheduling challenges caused by heterogeneous resources, multi-dimensional constraints and dynamic time-varying characteristics in 6G integrated satellite-terrestrial networks, this paper proposes an intelligent scheduling method based on a hierarchical collaborative architecture and unified heterogeneous graph representation. Firstly, a three-tier hierarchical collaborative architecture consisting of the satellite-borne layer, ground layer and edge layer is constructed, and multi-domain resources are abstracted into a heterogeneous graph. Secondly, after analyzing seven key constraints including latency, capacity and energy, the scheduling problem is formulated as a Constrained Markov Decision Process (CMDP). A lightweight Proximal Policy Optimization (PPO) algorithm integrating action masking and safety projection is designed to realize joint scheduling of spectrum, power, computing and storage resources. Simulation results demonstrate that compared with heuristic algorithms, two-stage optimization, discrete Deep Q-Network, round-robin and random allocation (lower-bound benchmarks), as well as vanilla PPO, the proposed method reduces the long-tail latency of high-priority services by 33%–58% and cuts the access collision probability of Internet of Things (IoT) devices by 56%. Meanwhile, it significantly improves the task completion rate and effective throughput within the handover window. From the engineering implementation perspective, structured pruning, low-rank decomposition and 8-bit integer quantization are adopted to compress the model parameters to one-third of the original size, enabling feasible millisecond-level inference deployment on satellite-borne platforms.  
      关键词：6G satellite-terrestrial integration;heterogeneous graph modeling;resource collaborative scheduling;lightweight deep reinforcement learning;proximal policy optimization   
      7
      |
      2
      |
      0
      <HTML>
      <L-PDF><WORD><XML><Meta-XML>
      <引用本文> <批量引用> 172868028 false
      更新时间：2026-09-29
    0