| The Internet of Things(Io T)integrates numerous ubiquitous,interconnected,and intelligent devices from the physical world through the internet,enabling various intelligent Io T applications.To support compute-intensive and latency-sensitive intelligent Io T applications,researchers have proposed a new computing network architecture called Mobile Edge Network(MEN),which extends remote cloud-based mobile computing.The core idea of MEN is to deploy computing,storage,and communication resources to the network edge to provide services nearby,ensuring the quality of service for users.Dynamic resource management technology,due to its efficient and flexible characteristics,has become a key technology for improving the performance of Mobile Edge Networks.However,due to the massive amount of high-value data involved in the applications served by MEN and the openness,mobility,and interaction of edge devices,the dynamic resource management process can easily lead to privacy leaks.Therefore,this paper focuses on the research of dynamic resource management technology with privacy protection for MEN,aiming to meet the requirements of both user service quality and privacy security.The main research content of this paper is summarized as follows:(1)A low-cost privacy-preserving dynamic resource management strategy is proposed.This strategy is aimed at a basic MEN system consisting of cloud servers,individual edge nodes,and multiple mobile terminal devices.The edge nodes manage resources to handle randomly arriving tasks,while privacy concerns arise from malicious attackers eavesdropping on their processing activities.Firstly,the dynamic resource management problem of the edge nodes is formulated as an optimization problem to minimize long-term cumulative costs and modeled as a Markov decision process model.Secondly,a Differentially Private Deep Qlearning based resource Management(DP-DQM)algorithm is proposed to solve this model while ensuring privacy.This algorithm introduces a noise mechanism satisfying differential privacy into the Deep Q-Network(DQN)and only adds noise to the decision output to reduce the cost of privacy protection.Additionally,the privacy guarantee and utility guarantee(learning error bound)of the DP-DQM algorithm are theoretically analyzed.Finally,simulation results verify that the proposed algorithm achieves comparable performance to deep Q-learning algorithms while ensuring privacy protection.(2)A globally privacy-preserving dynamic resource collaborative management strategy is proposed.For the complete MEN system consisting of a cloud server,multiple edge nodes,and multiple mobile terminal devices,the collaborative resource management of each edge node is considered to improve the overall Qo S of the system.However,due to privacy requirements,edge nodes cannot directly transmit user data to the cloud or other edge nodes.First,the dynamic resource management problem of multiple edge nodes is modeled as an optimization problem to minimize the average cumulative cost,and it is modeled as a Markov decision process model.Second,based on the federated learning framework and the DP-DQM algorithm,a Federated Differentially Private Deep Q-learning based resource Management(FDP-DQM)algorithm is proposed.The algorithm regards edge nodes as local clients using the DP-DQM algorithm to train a local model and then uploads the model to the cloud server for aggregation to obtain a public model with global features,which is then shared among all edge nodes to achieve resource collaboration management we ensuring global privacy security.In addition,the privacy guarantee and utility guarantee(learning error upper bound)of the FDP-DQM algorithm are theoretically analyzed.Finally,simulation results show that the FDPDQM algorithm can achieve equivalent performance to the CDP-DQM(DP-DQM based on centralized training)algorithm without considering collaborative privacy security while achieving global privacy protection. |
