Energy-Efficient Collaborative AI Service Migration and Resource Allocation Based on Traffic Prediction in 6G Networks

As a more advanced next-generation mobile communication system, sixth-generation (6G) will be tightly integrated with Artificial Intelligence (AI) technologies and will combine cloud-network-edge collaborative computing architecture to provide diverse AI services. However, due to the time-varying nature of network traffic, ensuring Quality of Service (QoS) and continuity for AI services remains a challenge. To address this issue, we focus on the migration of AI services in temporal dynamic traffic scenarios, and model it as an optimization problem. We aim at minimizing the long-term average system energy consumption, and propose a collaborative agent migration with deep neural network (DNN) deployment and dynamic bandwidth resource allocation algorithm based on traffic prediction (CAMD-TP). DLinear is used to predict the time-varying network traffic for AI service migration decisions. We apply Lyapunov optimization to convert the long-term stochastic optimization problem into a short-term deterministic problem. We propose an optimization algorithm based on the general Benders decomposition (GBD) technology to solve the AI service migration and dynamic bandwidth resource allocation problems. Experimental results show that, compared with two heuristic algorithms, CAMD-TP reduces the average system energy consumption up to 18.7%, and it outperforms existing algorithms in both traffic acceptance rate and throughput.