Research On Caching Strategy Based On Federated Learning In UAV-Assisted Mobile Edge Computing

With the development of the Internet of Things,the number of mobile devices and network data traffic has increased rapidly,placing a huge burden on backbone networks.To meet this challenge,Mobile Edge Computing(MEC)comes into being.Through the decentralization of cloud services,puts computing and storage resources to MEC servers close to users,thus relieving data transmission pressure on backbone networks and cloud data centers.However,the fixed architecture of MEC system has problems of poor flexibility and high deployment cost,and the application of MEC technology has limitations in hot spots and remote mountainous areas.In addition,when natural disasters occur,ground infrastructure can be destroyed,resulting in congestion or failure of communication networks.UAV-assisted MEC is an effective method to solve the above problems.Unmanned Aerial Vehicles(UAVs)have the advantages of low cost,high maneuverability,flexibility,good line-of-sight communication links,and ease of rapid on-demand deployment.The technology of using UAVs as air base stations to assist MEC can effectively reduce network congestion,expand network coverage in resource-scarce areas,and provide users with rapid emergency communications response and precise observation services.Moreover,the transmission of massive content in the network aggravates the pressure on the backbone network,and edge caching technology can effectively reduce network load and content transmission delay.In the UAV-assisted MEC environment,edge servers and UAVs can pre-cache the content frequently requested by users in the cloud data center,thus reducing the frequent communication with the cloud data center and the content transmission delay.However,because of the limited cache capacity,edge servers and UAVs cannot cache all the content requested by users.Therefore,how to achieve efficient content caching on edge servers and UAVs with limited caching resources,thereby reducing content transmission delay,has become an urgent problem to be solved.In addition,the effectiveness of caching strategy depends on the accuracy of user preference prediction to a large extent.Methods based on centralized training to predict user preferences need to transmit a large amount of local data to a central server for processing and training,which results in high communication costs and transmission delays.Federated learning has been widely used in the cooperative training of machine learning models among distributed devices.The user preference prediction method based on federated learning can improve the training efficiency and reduce the burden on the backbone network.However,the heterogeneity of user device resources,the limitation of computing resources and network communication resources usually lead to the problems of high training delay and energy consumption in federated learning.Therefore,it is of great theoretical and practical significance to study the optimization problem of user selection and resource allocation for federated learning and the problem of UAV-assisted edge caching based on user preference prediction.In view of the above problems,this paper carries out research from the following aspects:(1)Aiming at the problem of high training delay and energy consumption of federated learning is caused by the limited network communication resources and the heterogeneity of the user device resources during the federated learning training process,A federated learning-oriented optimization method for user selection and resource allocation is proposed in this paper.Firstly,the optimization problems of user selection and resource allocation for federated learning are analyzed.Secondly,the latency and energy consumption models of hierarchical federated learning are established respectively,and the optimization objective to minimize the training latency and energy consumption of federated learning is proposed.The optimization problem is formulated as a Markov Decision Process(MDP),and the global optimal user selection strategy and resource allocation strategy are solved through Double Deep Q Network Learning.Finally,DQN-USCRA,DDPG-CRA and Q-US are used as benchmark algorithms for experimental comparison with the algorithm proposed in this paper.The experimental results show that the algorithm proposed in this paper can effectively reduce the long-term training delay and energy consumption of federated learning.(2)In the UAV-assisted MEC environment,for the time-varying of user preferences,the limited network bandwidth resources and the limited cache capacity of UAV and edge base station cause the cache hit rate to drop and the content transmission time to extend,a UAV-assisted edge caching method based on user preference prediction is proposed in this paper.Firstly,the caching problem in the UAV-assisted mobile edge computing environment is analyzed.Secondly,a user preference prediction model based on federated learning is constructed,and an optimization objective to minimize the content transmission delay is established.The UAV-assisted edge caching optimization problem based on user preference prediction is formulated as MDP,and Dueling Deep Q Network Learning is used to solve the global optimal caching strategy and bandwidth resource allocation strategy.Finally,the proposed algorithm is compared with LFU,CCAZD and TLCA.The experimental results show that the proposed algorithm can effectively reduce the content transmission delay and improve the cache hit rate.