LAUSR

In the complex dynamic environment of the Internet of Things (IoT), the massive terminal devices generating stochastic network service requests pose significant challenges to service function chain (SFC) deployment, particularly in dynamic configuration and resource utilization efficiency. To address these issues, this article proposes an SFC deployment algorithm incorporating incremental learning with resource-aware mechanisms. First, an incremental learning-based deep-width neural network model is designed, which integrates the spatiotemporal feature extraction capabilities of graph convolutional networks (GCNs) and temporal convolutional networks (TCNs) with the continuous learning ability of the broad learning system (BLS), enabling dynamic prediction of virtual network function (VNF) resource demands. Then, based on the prediction results, resource capacity awareness is used to assess the VNF deployment potential of nodes. Under the constraints of delay and network resources, an optimization problem is formulated with the objectives of maximizing the benefit-cost ratio of SFC deployment and minimizing deployment energy consumption. Finally, to solve this optimization problem, a multiagent generative adversarial imitation learning algorithm is proposed, which guides the agents’ policy learning through expert experience to improve algorithm performance. Simulation results show that the proposed algorithm performs well in terms of enhancing node resource awareness, SFC deployment benefits, and acceptance rates.