Research On Cooperative Offloading Technology For Services In Power Internet Of Things

With the construction and development of the power Internet of Things(IoT),the services in power IoT such as line detection,remote monitoring,smart home,and remote meter reading are diversified.At the same time,the number of terminal connections has grown rapidly,followed by an explosion of network data,service latency has become more and more severe,and system energy consumption has also increased.Traditional cloud computing has obvious advantages in processing computing-intensive tasks because of its abundant computing resources,but it is usually far away from the terminal device,which makes the task response time longer.Edge computing complements the advantages of cloud computing because it is closer to the terminal device and has certain task processing capabilities.Therefore,it is particularly important to perform effective task offloading and resource allocation at the edge layer and cloud layer.In the cloud-edge architecture of power IoT,collaboration between edge IoT agents on the horizontal structure,as well as collaboration between edge IoT agents and cloud center on the vertical structure,can alleviate the problem of insufficient capacity of a single node and meet the resource requirements of a large number of services.Based on the processing capability of different compute nodes,nodes cooperate with each other to fully improve network resource utilization,effectively reduce task delay and system power consumption,and improve service quality.For the collaborative offloading of Power IoT services between edge IoT agents,many methods have been discussed,and a large part of them focus on task delay and energy consumption,without considering the differentiated needs of different services.There is no refinement task.In this paper,the resource block pre-allocation stage is added,and the edge side benefit,service level and number of bids are considered jointly,and the resources of the edge IoT agent are logically divided into multiple resource blocks to serve the Power IoT service of the corresponding level.In addition,the current mainstream methods do not consider the service motivation of resource providers from the perspective of future marketization.In this regard,the paper combines economic models to describe the service relationship between edge IoT agents and Power IoT services,and proposes an improved multi-round combination based on The edge collaborative offloading method of auction,which performs dynamic pricing based on the urgency of the service,and comprehensively considers the available resource capacity of the edge IoT agent and its distance to the Power IoT services for bidding.Finally,the performance of the method proposed in the paper is evaluated by simulation experiments,which improves the benefit of the edge side and effectively reduces the service delay and system energy consumption.For the cloud-edge collaborative offloading problem in the vertical structure,most of the current literature is the overall offloading.The paper takes the power video service as the starting point,and studies the offloading problem of the associated sub-tasks in the vertical structure of the power IoT.The associated subtasks are modeled as a directed acyclic graph,and Graph Neural Network(GNN)is used to extract information about its structure,and then combined with the multi-agent deep reinforcement learning algorithm,the cloud center and each edge IoT agent are regarded as multi-agents.The cloud center with global perspective is used as the global experience replay buffer.And a cloudedge cooperative offloading method based on MADRL assisted by GNN is proposed.Finally,simulation experiments are carried out to verify the actual performance of the algorithm,mainly for power video services.The average delay is used as the evaluation standard to measure the performance of the algorithm.The evaluation is made by adjusting the number of training times,the density of the DAG structure diagram,and the number of edge IoT agents.The results show that this method effectively reduces the delay of power video services and improves its performance service quality.