DRL-Based SFC Deployment in Time-Aware Aerial Computing Systems

The Space-Air-Ground Integrated Network (SAGIN), as a key enabler of aerial computing, offers cross-domain coordination and extensive coverage, making it a promising architecture for sixth-generation (6G) communication systems. Deploying Service Function Chains (SFCs) in SAGIN further enables agile, low-latency, and resilient network services. However, dynamic topologies, constrained resources, and stringent latency and reliability requirements pose significant challenges to its effectiveness. To tackle these challenges, we introduce a cost-aware Time-Expanded Graph (C-eTEG) framework that models temporal variations in network capacity and resource costs. Leveraging this model, we formulate a joint optimization problem for SFC deployment and routing, with the objective of maximizing long-term service profit. A Dueling Double Deep Q-Network (D3QN)-based algorithm is developed to enable efficient and robust decision-making, further strengthened by adversarial reinforcement learning (ARL) to enhance resilience under worst-case conditions. Simulation results demonstrate that our method significantly improves resource utilization and service success rates, outperforming existing reinforcement learning baselines in dynamic SAGIN environments.