Research On Efficient Resource Management Based On Edge Computing

With the flourishing development of artificial intelligence and the Internet of Things(Io T),wireless networks are evolving towards deep integration of perception,communication,computation,and intelligence,enabling the efficient,intelligent interconnection of people,machines,things,applications,and data services.However,the diverse demands of intelligent applications in wireless networks are continuously growing.The current wireless network architecture cannot meet latency,high energy efficiency,privacy,and security requirements.In this context,edge computing and edge intelligence technologies have emerged as effective technical solutions.However,to fully leverage the potential of wireless edge computing,many challenges need to be addressed in the dynamic management of resources,distributed intelligent learning,and data privacy protection.Collaborative computing aims to fully utilize the data and resources of end devices,edge nodes,and the cloud,to improve the efficiency,reliability,and security of wireless networks and provide technical approaches to address these challenges.Therefore,this dissertation,combined with collaborative computing technology,focuses on the critical challenges of wireless edge networks based on wireless edge computing.The specific content and innovation points are summarized as follows:1.This dissertation proposes a novel edge-end collaborative computing network architecture to meet the low-latency processing requirements of industrial Internet of Things(IIo T)tasks and designs efficient and diverse task offloading solutions between resource-constrained IIo T devices and edge servers.Firstly,based on the edge-end collaborative computing mode,a system latency and energy consumption model is established.Secondly,an importance-aware task partition strategy is proposed to meet the different service quality requirements of heterogeneous and dynamic tasks.To further encourage IIo T devices and edge services to participate in resource sharing and collaboration,an online auction mechanism is designed.Then,with the goal of maximizing system utility,a joint task offloading and resource allocation optimization method is formed.Finally,an online incentive algorithm is proposed to obtain optimal task execution methods and price update strategies,achieving faster task execution efficiency and higher resource utilization.Theoretical and numerical results demonstrate that the proposed algorithm maximizes system utility while ensuring incentive compatibility,individual rationality,computational efficiency,and feasibility.2.Aiming at the challenges of heterogeneous resource management and diverse Io T application requirements,a Cybertwin based end-edge-cloud network architecture is proposed,and a hierarchical collaborative computing scheme toward cloud,edge servers,and end devices is researched.Firstly,collaborative edge computing offloading and hybrid alternate offloading modes are designed for delay-tolerant and delay-sensitive missions,respectively,to provide users with high experience quality,low latency,and ultra-reliable services.Secondly,a communication,computation,and caching system model is established,and a joint hierarchical task offloading and multidimensional resource allocation optimization problem is formulated to maximize system processing efficiency.Based on this,four task execution scenarios are analyzed under the two offloading modes dominated by the edge.Then,the optimization problem is transformed into a Markov decision process,and a multi-agent deep deterministic policy gradient algorithm is designed.A flexible task offloading and resource management strategy is obtained through distributed privacy-preserving model training methods.Numerical results demonstrate that the proposed algorithm enhances the task completion ratio while ensuring lower system overhead relative to the deep deterministic policy gradient algorithm.3.To address the issues of system heterogeneity and dynamic changes in multi-layer computing networks,this dissertation proposes an optimization scheme for scheduling and deploying large-scale computing federated learning tasks,aiming to employ cloudfog-edge-end collaborative computing technology to accelerate the federated learning training process.Firstly,a novel adaptive training and aggregation federated learning framework is proposed,in which local models can be trained on end devices,edge nodes,and fog nodes,and the global aggregator can be flexibly configured in the edge,fog,and cloud tiers.Secondly,based on the multi-layer FL distributed training mechanism,a delay and energy consumption model is established for the training and aggregation process,and an optimization problem of joint training,aggregation node selection,and resource allocation is formulated.A twin delayed deep deterministic policy gradient algorithm is proposed,which integrates digital twins into the multi-tier computing network to assist in perceiving system state information and can quickly obtain the optimal node selection and resource allocation strategy.Finally,performance evaluation is conducted based on a developed experimental prototype,the results show that the proposed algorithm effectively reduces the average system delay and energy consumption while improving accuracy compared with traditional federated learning,federated edge learning,random policies,and deep deterministic policy gradient algorithms.4.An optimized federated learning model training scheme for cloud-edge collaboration is investigated to address the statistical heterogeneity issue in edge computing systems.Firstly,the impact of Non-iid data on the accuracy and communication overhead of hierarchical federated learning models is analyzed.Based on this,a selectively hierarchical federated learning framework is proposed,which achieves intelligent training with high accuracy and low communication cost through the collaborative interaction of end devices,edge servers,and the cloud.Secondly,a joint data selection and communication resource allocation strategy is designed based on the soft actor-critic reinforcement learning algorithm to support efficient scheduling of end devices and minimize communication costs and learning evaluation losses.In addition,local model drift correction and hierarchical synchronized update mechanisms are adopted to accelerate convergence.Simulation results show that the proposed algorithm improves training accuracy and communication efficiency while satisfying data privacy.