物联网学报 (Mar 2024)
Evaluation and optimization of carbon emission for federal edge intelligence network
Abstract
In recent years, the continuous evolution of communication technology has led to a significant increase in energy consumption.With the widespread application and deep deployment of artificial intelligence (AI) technology and algorithms in telecommunication networks, the network architecture and technological evolution of network intelligent will pose even more severe challenges to the energy efficiency and emission reduction of future 6G.Federated edge intelligence (FEI), based on edge computing and distributed federated machine learning, has been widely acknowledged as one of the key pathway for implementing network native intelligence.However, evaluating and optimizing the comprehensive carbon emissions of federated edge intelligence networks remains a significant challenge.To address this issue, a framework and a method for assessing the carbon emissions of federated edge intelligence networks were proposed.Subsequently, three carbon emission optimization schemes for FEI networks were presented, including dynamic energy trading (DET), dynamic task allocation (DTA), and dynamic energy trading and task allocation (DETA).Finally, by utilizing a simulation network built on real hardware and employing real-world carbon intensity datasets, FEI networks lifecycle carbon emission experiments were conducted.The experimental results demonstrate that all three optimization schemes significantly reduce the carbon emissions of FEI networks under different scenarios and constraints.This provides a basis for the sustainable development of next-generation intelligent communication networks and the realization of low-carbon 6G networks.
