Adaptive Service Function Chain Orchestration via DyFLO for IoT-Enabled Edge-Computing-Enhanced Space-Air-Ground Network

Internet of things enabled (IoT-Enabled) Edge-computing-enhanced space–air–ground integrated network (EC-SAGIN) is critical for smart cities. However, effective service function chain (SFC) orchestration faces three core challenges. The first is the heterogeneous service challenge, driven by diverse service requirements and quality of service (QoS) constraints. The second is the dynamic topology challenge, which arises from the extreme topological dynamics of the integrated network. The third is the long-term maintenance challenge, rooted in inefficient virtual network function (VNF) sharing and inadequate full-lifecycle management. Traditional static methods suffer from low resource utilization, poor adaptability, and compromised service continuity. To address these challenges, this paper proposes DyFLO (Dynamic Fusion-driven Lifecycle Optimization), an adaptive SFC orchestration framework with three core innovations. The first innovation is a time-aggregated graph (TAG) that dynamically models the spatiotemporal evolution of the network. The second innovation is a hybrid mechanism combining graph attention network (GAT) and gated recurrent unit (GRU), which effectively mines topological correlations and VNF sequence dependencies. The third innovation is an actor-critic deep reinforcement learning (DRL) model that learns optimal deployment and migration strategies. This DRL model is further augmented by a pseudo-VNF sharing mechanism and a full-lifecycle optimization mechanism to enhance practical applicability. Experimental results show that the framework outperforms representative algorithms in key performance metrics. It achieves 35.3% higher system revenue than the Lyapunov optimization-based algorithm TMSM and 106.6% higher than the meta-heuristic particle swarm optimization (PSO) algorithm. Its effective service capability also stands out, exceeding the TMSM algorithm by 53.7% and the PSO algorithm by 80.4%. Beyond these comparative advantages, the framework demonstrates strong adaptability to varying SFC lengths, resource abundances, and network loads. It thus provides a feasible and efficient SFC orchestration solution for EC-SAGIN in smart cities.