Application of a PPO-Based Scheduling Algorithm with Multi-Dimensional Attention Mechanisms in Satellite Edge Computing

With the rapid development of Low Earth Orbit (LEO) satellite networks, satellite edge computing has emerged as a key technology for enabling integrated space-air-ground intelligent services. By deploying computational resources onboard satellites, it supports real-time task processing and intelligent scheduling for ground users. However, the dynamic network topology, limited onboard resources, and bursty workloads challenge traditional heuristic scheduling methods. This paper proposes an enhanced Proximal Policy Optimization (PPO) algorithm incorporating multi-dimensional attention mechanisms. The algorithm is built on an Actor-Critic framework with a multi-layer Transformer architecture, where each layer integrates multi-scale attention, task-resource cross-attention, and adaptive sparse attention. These mechanisms are fused using learnable residual weights to enhance policy expressiveness and generalization. Experimental results demonstrate that the proposed method significantly improves task completion rates while maintaining low latency and energy consumption, showing strong potential for satellite edge computing applications.