Research On Learning-Based Content Caching Strategy Of Fog Node

Fog Radio Access Network(F-RAN)is one of the 5G-oriented mobile network architectures.It uses the computing and storage functions of the fog node in the fog layer to improve network performance gains.In recent years,with the rapid development of related technologies of smart mobile terminals,the video stream on the edge network will increase substantially and occupy most of the mobile data traffic.Moreover,there exists a large number of Io T applications requirements in the 5G era,the number of connections and traffic density will increase significantly,and cloud servers will also have a lot of pressure on computing and processing.In order to support the application of the Internet of Things,the Fog radio access network as a solution has caused extensive research in the industry.F-RAN makes full use of the computing and caching capabilities of mobile edge devices,and it proposes to sink computing power to where close to users,pre-cache popular content on local fog nodes during off-peak periods,which in case that a large number of content requests emerge during peak periods.Because the fog nodes equipped at the mobile edge are densely deployed but the cache space of each one is limited,it is very meaningful to study the problem of fog node content caching.The key to the content caching problem of fog nodes is that the popularity of content in real life is unknown and constantly changing,and the users served by fog nodes are also uncertain.Based on this premise,with the goal of maximizing caching utilization efficiency and improving user experience,how to design caching and updating decisions for fog nodes is an urgent challenge to be solved.This paper focuses on the problem of user clustering and fog node caching in the fog layer and terminal layer.The main research content and innovation work are summarized as follows:Firstly,in the network based on the F-RAN architecture,in order to improve the caching utilization of the fog node,this paper proposes to model the user clustering problem as a0-1 knapsack problem with knapsack dependent item values.This model aims to maximize the degree of similarity between users and the fog nodes,and it also propose two kinds of clustering algorithms,corresponding to solve the exact solution and approximate solution respectively.Secondly,after selecting a appropriate fog node for each user based on the above clustering algorithm,this paper studies the content caching problem of fog nodes.This paper proposes to model the content caching problem of the fog node as a multi-arm bandit problem.From the perspective of the user,the data transmission rate of the content obtained from a fog node is taken as the cache caching of the fog node.The model aims to maximize the cache reward,and then the caching algorithm based on reinforcement learning determines exactly the caching decision for the fog node.Finally,in real life,the rule of user movement are complex,and it is difficult for each fog node to accurately predict the popularity of each content.Therefore,combining the user clustering algorithm proposed in this article with the fog node caching algorithm based on reinforcement learning can improve caching performance of the fog network system.In the simulation,this paper performs user clustering in a large-scale time range,meanwhile in a small time-scale range,caches and updates for every fog nodes in the fog layer based on the clustering results.Taking into account the user's movement,the simulation scene is divided into static and dynamic parts.Finally,the simulation results show that with the number of clustering increases,the caching hit ratio gradually increases and then stabilizes.The positive feedback effect of clustering on the caching algorithm is verified,indicating that the clustering algorithm and caching algorithm proposed in this paper can improve the caching performance of the fog network and improve the user experience.