Research On Resource Allocation Algorithm For Federated Learning In Internet Of Vehicles

With the rapid development of IoV,vehicles use more and more advanced sensing equipment and computing power to collect and process a large number of datasets.In order to protect user privacy while training models with multiple vehicle datasets,federated learning is introduced into the IoV.Due to the heterogeneity of different vehicles in datasets,computing resources and wireless resources,different vehicle selection and resource allocation schenes have a great impact on federated learning training process.The topic of the thesis is from Beijing Natural Science Foundation Project "Research on Internet of Vehicles Resource Allocation Algorithm Based on Video Content Understanding Drived by Dynamic Spatiotem-poral Data"(Project No.:4202049).The thesis proposes two new algorithms:Content-based vehicle selection and resource allocation for federated learning in IoV and Resource allocation algorithm based on resource utility rate maximization,they respectively to realize the global model precision and the improvement of resource utility,which is suitable for IoV.The main research contents of this thesis are as follows:(1)This thesis systematically summarizes the related studies on federated learning technology in the Internet of Vehicles.Firstly,the architecture and key technologies of Internet of Vehicles are briefly summarized.Then it focuses on the research status of federated learning resource management algorithm in Internet of Vehicles.In view of he shortcomings of the existing resource allocation algorithms,the research direction of this thesis is put forward,which lays a foundation for the research content of this thesis.(2)Aiming at the heterogeneous problem of imbalanced distribution of data in the vehicle dataset in the Internet of Vehicles,this thesis proposes a vehicle selection and resource allocation algorithm based on content balance.Compared with the existing algorithms,this thesis jointly considers the influence of imbalanced dataset content distribution,vehicle computing capacity and wireless channel,designs the vehicle user selection and wireless resource allocation as an optimization problem,and solves the optimization problem by introducing genetic algorithm.Theoretical analysis and experimental results show that the proposed algorithm can not only improve the accuracy of the model,but also accelerate the convergence speed of the model.(3)In order to solve the problem that the utility rates of global computing resources and communication resources of the vehicle terminal and server are ignored by the existing algorithms under different resource pricing scenarios which leads to low global resource utility rates,this thesis proposes a resource allocation algorithm based on the maximization of resource utility rates.First of all,the gain from the performance improvement of federated learning model is reasonably quantified,and then the resource cost of computing and communication processes is considered jointly.The optimization function is constructed by maximizing the global utility derived from the global model payoff per unit time minus the total computing cost and communication cost.Compared with the existing algorithms,this thesis improves the global resource utilization of federated learning by allocating computing and wireless resources jointly.Theoretical analysis and experimental results show that the proposed algorithm can find the optimal resource allocation method under different computing resource and wireless resource pricing scenarios,maximize the global utility rate,and ensure the effective training of the global model.