Research And Application Of Edge Intelligent Collaboration Strategy Based On Federated Learning

With the widespread use of the Internet of Things and edge devices,the center for processing data will be transferred from the cloud center to the edge network,and the method of federated learning has also been created and widely used.Using federated learning distributed training artificial intelligence model in the edge network has become a research hot spot and main research direction in the field of edge intelligence.At present,the research and application of the federated learning method in the edge network has achieved many results,but there are still the following problems: first,the heterogeneity of edge devices leads to the low accuracy of the federated learning model;second,the local data heterogeneity of edge devices leads to the low federated learning and training efficiency.This thesis conducts in-depth research on the above-mentioned problems,and the research results are as follows.Aiming at the problem of low model accuracy when edge heterogeneous devices participate in federated learning,this thesis introduces an integrated knowledge distillation method,designs an intelligent collaboration method for federated learning based on integrated distillation called Fed EKD.Firstly,this thesis analyzes the influence of heterogeneous devices on federated learning under the environment of limited edge network resources,and then constructs a new federated learning architecture based on the analysis.Secondly,this thesis designs a grouped local training mode to carry out the feature migration of local data through the integrated knowledge distillation method.The server performs federated aggregation after the local integrated distillation training is completed.Finally,compared with other baseline models in terms of global model accuracy and convergence time,the effectiveness of the intelligent collaboration method Fed EKD is verified.Aiming at the low training efficiency of the federated learning intelligent collaboration method based on integrated distillation due to the heterogeneity of local data of each edge device in Fed EKD,this thesis analyzes the calculation process,introduces deep reinforcement learning method,and designs an intelligent cooperative grouping device selection algorithm DRL-Fed GDS based on deep reinforcement learning.Firstly,by modeling and analyzing the equipment grouping and training situation in Fed EKD,this thesis designs an optimization model to solve the large difference in group training time caused by data heterogeneity.Secondly,this thesis designs grouping strategies for deep reinforcement learning agents combined with deep reinforcement learning models.Finally,the difference between the maximum and minimum values of the group training time are compared with other baseline models to verify the effectiveness of the group device selection algorithm DRL-Fed GDS.According to the actual needs of the background project,the application visualization system of the edge intelligent collaboration algorithm based on federated learning is designed and implemented.The federated learning intelligent collaboration method based on integrated distillation and the intelligent collaborative grouping equipment selection method based on deep reinforcement learning are applied to the application visualization system,and a set of edge intelligent collaboration system is constructed,which provides a specific solution for improving the intelligence level of industrial manufacturing.