Research On Mobile Node Cooperation Strategy For Hierarchical Federated Learnin

With the rapid development of 5G wireless communication technology,the number of intelligent terminal devices worldwide has increased dramatically.Multiple edge nodes cooperate to perform computing tasks in order to meet the service needs of multiple terminal users and effectively reduce processing latency.For hierarchical federated learning,this article focuses on issues such as how to design the offline node cooperation strategy,how to design the online node cooperation strategy,and how to select appropriate cooperation nodes based on reputation values.Thus,achieving the goal of reducing latency and improving overall system performance.Firstly,in order to reduce the latency of parameter processing in application scenarios with low data privacy sensitivity,we design an efficient offline node cooperation strategy.The problem of minimizing delay has been proven to be NP-hard.We propose a Dynamic Cooperative Cluster Algorithm(DCCA).The design of DCCA algorithm is divided into two steps.For the first step,we propose an initial dynamic cooperative cluster algorithm based on similarity.For the second step,based on the computing capacity and transmission capacity of the nodes,another algorithm is proposed to adjust the dynamic cooperative cluster according to the core nodes.The experimental results show that compared with existing algorithms DOA and Fed Avg,the maximum delay of DCCA algorithm is reduced by 12.44% and 18.70%,respectively.Secondly,in order to accurately characterize the real-time state of hierarchical federated learning networks and whether nodes collaborate,an efficient online node cooperation strategy is designed.An online algorithm for node cooperation based on Lyapunov optimization theory(ONCA)is proposed for the problem of total cost minimization.In ONCA,the cooperation among nodes can be adjusted adaptively according to the dynamics of system state.Nodes prefer to select neighbors with high computing and transmission capabilities to cooperate,so that their capabilities can be fully utilized.The experimental results show that compared with existing algorithms Hier FAVG and Fed Avg,the average cost of ONCA algorithm is reduced by 13.86% and 18.04%,respectively.Finally,a reputation based cooperative node selection strategy is proposed to address the issue of how to select the suitable cooperative nodes for hierarchical federated learning.In order to optimize the accuracy of node selection,an algorithm of node selection based on reputation model(NSRA)is proposed.In NSRA,the edge server selects the node with high reputation prediction value to participate in the model training,and the node selects the neighbor node with high transmission capacity to cooperate.Thus,NSRA can optimize the trust between nodes and continuously improve the reliability of cooperative nodes.The experimental results show that the accuracy of the NSRA algorithm is improved by 19.38%compared to the existing algorithm RSA.