| With the rapid development and convergence of networks such as Io T,users’ demand for innovative,technological,and high-quality services is increasing every day.The using of these services often implies the emergence of massive computing volume and harsh latency requirements.Mobile Edge Computing(MEC)can alleviate the latency and congestion caused to users by massive computing demands in cloud computing.But when a large number of users with computing demands flock to mobile edge networks,the contradiction between users’ diverse and high-level service demands and limited computing resources becomes increasingly apparent.Therefore,in order to cope with the ever-increasing number of access users and the explosion of data traffic,it is necessary to fully exploit the potential network resources and effectively utilize the scattered and limited resources in the edge computing network.This paper investigates the task offloading and resource allocation problem of mobile edge computing in the converged network scenario of mobile edge computing with ultra-dense heterogeneous networks and vehicular networking by focusing on the diverse characteristics of tasks,fully exploiting the available free resources in the network,and using performance metrics such as latency and task success rate.The following is a summary of the main work and innovation points of this paper.First,in the traditional edge network,in response to the problem of high response latency and low task success rate caused by the contradiction between the large number of users,differential user demand and limited distributed resources,this paper constructs an edge computing network architecture incorporating Software Defined Network(SDN)technology,setting up an SDN controller to generate task offloading order decisions by obtaining global The SDN controller is set to generate task offloading order decisions by obtaining global information using reinforcement learning algorithms.An improved Q-learning based on fuzzy logic algorithm is proposed to simplify the complex spatial states by constructing Fuzzy Inference Systems(FIS)to generate the offloading order,select the server with the current maximum utility value for offloading according to the offloading order,and use the free user resources capable of D2 D communication as the extension of network resources.In addition,the offload order decision proposed in this paper is compared with min-min,max-min and random offload order in terms of average arithmetic power metrics and success rate metrics based on task data and time delay,and the effectiveness of the algorithm in improving the average arithmetic power and success rate obtained by users is demonstrated.Second,to address the contradiction of the limited resources of fixed servers and the high cost of deploying servers in traditional edge networks,this paper constructs a converged network model of vehicular and edge computing networks,and proposes a task offloading decision based on deep reinforcement learning for the offloading problem of two tasks with typical characteristics,namely,computationally intensive and delay-sensitive.In this part of the study,to solve the allocation problem of vehicle resources and ensure that vehicles can provide quality services,a scheme of collaborative vehicle computing is proposed considering factors such as limited computational capacity of individual vehicles and variable dwell probability.A Weighted Clustering Algorithm(WCA)based on relative distance,number of vehicles and dwell probability is proposed for vehicle clustering services.Finally,deep reinforcement learning is used to generate offloading decisions and perform experimental simulations.Different experimental environments are set up in the simulation to compare with DQN-FES algorithm and Max S algorithm with average user satisfaction as the performance index. |
