Research On Key Technologies Of Internet Of Things Data Value Sharing Based On Federated Learning

With the rapid development of technology,the data in the Internet of things is growing explosively,and people’s demand for data value sharing of the Internet of things is becoming higher and higher.At present,using federated learning technology to share model parameters instead of directly sharing private data was considered safely and efficiently.The goal of data value sharing of the Internet of things was achieved by it.However,the computing power level of computing power nodes in the Internet of things varies greatly.Most of the data in the Internet of things are heterogeneous data.If they directly participate in the federated learning process,the training efficiency of the federated learning model will be low,and the accuracy of the federal learning aggregation model will be low.In other words,actual use needs will not be met.The selection of federated learning training nodes and federated learning aggregation algorithm were studied in this paper to improve the efficiency of data value sharing in the Internet of things.Firstly,aiming at the problem that the training nodes with great differences in computing power level in the Internet of things are assigned to the same federated learning task,which leads to the low training efficiency of the federated learning model,a federated learning computing power resource allocation method for concurrent tasks is proposed.By studying the federated learning node selection technology and combined with the distributed state feedback strategy,a federated learning training node selection mechanism based on distributed state feedback is advanced.Through simulation experiments,the proportion of high-performance computing nodes participating in tasks and the time-consuming of completing all federated learning tasks are compared.It is verified that the proposed method can shorten the time-consuming of Federated learning tasks and improve the efficiency of data sharing in the Internet of things.Secondly,aiming at the low accuracy of Federated learning aggregation model caused by the direct aggregation of local models for heterogeneous data training,a method to improve the accuracy of Federated learning model for heterogeneous data is raised.By studying the federated learning aggregation algorithm,combined with information entropy computer system,multi-stage hierarchical aggregation strategy and time weighted fading strategy,an asynchronous federated learning aggregation algorithm based on multi-stage hybrid weighting is proposed.By comparing the accuracy of the federated learning model with the traditional federated average aggregation algorithm,it is verified that the proposed method can improve the accuracy of the federated learning model and realize efficient data sharing in the Internet of things.