Research On Energy-efficient Distributed Edge Computing Based On Federated Learning

With the vigorous development of 5G-related technologies,the data generated at the edge of the network has shown explosive growth.Gathering these data for centralized processing will cause a lot of energy consumption,which is a huge challenge for the traditional cloud computing architecture.In this case,allowing data to be directly processed at the edge of the network is a reliable solution,and edge computing emerges as the times require.Among the many distributed edge computing methods,the method based on Federated Learning(FL)has attracted people's attention.It can cooperate with multiple edge servers to train machine learning models to protect user data privacy while providing users with intelligent services.We call this architecture Federated Edge Intelligence(FEI).Although the FEI system avoids the energy consumption of transmitting data to the cloud,the optimization of the energy consumption of the distributed training of the ML model on the edge server is still an urgent problem to be solved.There have been some studies on the energy consumption and time of each working node in the FL model training process,however,these works have not analyzed the overall energy consumption of the system from the perspective of the system as a whole,considering the whole process from data upload to completion of training.In addition,there is currently a lack of work related to the measurement of real energy consumption data on actual equipment.Therefore,the problem we need to solve is very difficult and challenging.In order to deal with these problems,this paper proposes an energy-efficient FEI(EEFEI)framework,which introduces a modeling analysis of FEI's overall energy consumption expenditure.The main contributions of this paper are as follows: We first analyze and model the energy consumption of the FL model in the FEI system,and quantify the relationship between the energy consumption of the system and the key parameters in the training process.Secondly,we analyzed the energy consumption optimization problem of the proposed EE-FEI system,proved the biconvex characteristics of the key parameters of the problem,and used the Alternate Convex Search(ACS)algorithm to solve the problem.Finally,we conducted a large number of real data experimental analysis through the built hardware platform,and evaluated our theoretical results.The final experimental results show that EE-FEI can greatly reduce the energy consumption of the FEI system,up to 49.8%.Our proposed method can indeed optimize the overall energy consumption of the system while ensuring the accuracy of the training model.