Research On Mobility Model Based On Federated Learning And Its Application In Edge Computing

With the advent and popularization of 5G,the interconnection of everything in real life scenarios,and the rapid development of technology have brought many highquality services and applications,such as autonomous driving,smart wearable devices,and smart homes.These services not only provide users with a convenient lifestyle,but also satisfy users' desire for high-quality experience.However,at the same time,high-quality services have produced exponentially increasing data volume,and network transmission delays and other failures have appeared,and there are some urgent problems that need to be alleviated.In recent years,according to data analysis,due to network delays and other failures,traffic safety problems caused by untimely or incomplete data transmission are widespread.Among them,traffic accidents and traffic jams have become one of the main factors that endanger people's lives and property.In order to alleviate the traffic problems caused by network delays,many of the existing studies propose mobile models for specific environments and predict network or mobile behaviors,and improve the accuracy of predictions by proposing improved methods to the models,while ignoring them The influence of data distribution on the parallel training of the model and the improvement of prediction accuracy.In response to the above mentioned problems,this paper proposes a mobility model based on federated learning,which uses feature-based data deployment methods to improve network and mobility prediction accuracy.Use mobile models to simulate mobile scenarios in actual applications to study traffic safety issues,predict the network quality in a short period of time for mobile users who establish connections,adjust the network conditions in advance.The following is the main research: This article first describes the research background and significance of mobile models and federated learning,and introduces the classification of mobile models and the related theoretical knowledge of federated learning.In the environment of edge computing,according to the mobility of users,information is transmitted between users in the network,establish a mobile model and a network model.In order to adapt to the network structure of edge computing,a system model composed of a server and multiple clients was created.By randomly selecting several clients that meet the requirements,the federated learning method uses multiple nodes to learn in parallel to improve learning efficiency.In order to further improve the speed of model training and the accuracy of prediction,a feature-based data deployment method is proposed based on the mobile model.And apply SVM algorithm and LSTM algorithm respectively to compare different data deployment methods to verify the effectiveness and wide applicability of data deployment to improve the accuracy of mobile model prediction.On the basis of the results,the mobile model based on federated learning proposed in this paper can improve the accuracy of prediction through reasonable data deployment,and it will also help to improve the stability and convergence speed.By comparing different data deployment methods,it is proved that under the same experimental environment and equipment,feature-based data deployment can significantly improve the accuracy of prediction and the speed of parallel training.