Satellite Edge Intelligent Computing Network: A Resource Management Method Based on Multi-Agent Deep Reinforcement Learning

The integration of artificial intelligence with satellite edge computing (SEC) networks presents a transformative solution for time-sensitive emergency remote sensing applications. However, it encounters critical challenges in resource-constrained low Earth orbit (LEO) environments. This paper proposes a novel multi-agent deep reinforcement learning framework to address the distributed resource management problem in dynamic SEC networks. Facing urgent remote sensing processing tasks, we establish a fully cooperative satellite edge intelligent transmission and computing architecture, in which each LEO satellite acts independently as an autonomous agent based on local observations. In order to jointly optimize the task offloading ratio, i.e. the corresponding allocated communication and computing resources of distributed multi-satellites, we propose a multi-agent proximal policy optimization algorithm with an actor-critic and centralized training-distributed execution framework. This approach implements the adaptive policy update by the centralized critic and the on-board real-time decision-making by the distributed actors in stages, significantly improving the efficiency of strategy learning and execution. The simulation results demonstrate the proposed scheme effectively reduces task execution time and satellite energy consumption during multi-satellite collaborative transmission and computation, and outperforms the benchmark algorithms. This verifies that the solution is able to balance mission-critical quality of service requirements and sustainable resource utilization in emergency situations, and is expected to achieve space-edge intelligence for the next-generation satellite-ground integrated network.