Heterogeneous Resource Allocation in Space-Air-Ground-Integrated Networks: A Multi-Agent Deep Deterministic Policy Gradient Approach

In recent years, space-air-ground integrated networks (SAGIN) have attracted considerable attention for their potential to support various applications in future 6G systems. However, the diversity of tasks generated by different network participants and the heterogeneous distribution of resources in SAGIN pose significant challenges in matching task demands with available resources. To tackle this issue, we propose a multi-agent deep deterministic policy gradient (MADDPG) algorithm with centralized training and decentralized execution (CTDE) that jointly addresses offloading decisions and heterogeneous resource allocation. Tasks are categorized into three distinct types with task priorities considered. Low Earth orbit (LEO) satellites equipped with edge computing capabilities are modeled as agents. Each agent makes decisions regarding heterogeneous resource allocation and offloading based on its own observations, while global action and state information are used to update deep neural networks. The CTDE-MADDPG algorithm enables low-latency distributed decision making in complex networks without the need for time-consuming state synchronization and centralized decision making. The simulation results demonstrate that the proposed scheme effectively minimizes energy consumption while ensuring tasks are completed within latency and resource constraints, achieving superior performance compared to the baseline algorithms.