Abstract
Edge Computing is an optimistic technology that can extend the necessary support for vehicular applications. In this paper, an effective edge-computing framework is developed to improvise task scheduling. A task partition and scheduling algorithm are developed to decide the workload allocation and schedule the execution order of tasks offloaded. Then, according to the characteristics of task scheduling, design the corresponding state-action space and reward function; and finally, taking into consideration the complexity of task scheduling and computing resource allocation, the pointer network is trained by multi-agent fuzzy deep reinforcement learning; this allows the pointer network to account for the dynamic nature of During the process of network fusion, it is used to find a solution for the issue of weight distribution for each agent.
The simulation showcases that the proposed method is superior. Furthermore, it has significant advantages in terms of convergence speed and optimal performance. It has a high degree of flexibility in the ever-changing and intricate electromagnetic environment. The capabilities of the Internet of Vehicles' job offloading system have been significantly increased because of this improvement.
It is widely believed that the Internet of Vehicles (IoV), which incorporates cutting-edge technologies such as connectivity, big data, and artificial intelligence, will play a significant role in the development of the next-generation intelligent transportation system. In recent years, the Internet of Vehicles has given rise to a significant number of novel computer jobs, such as augmented reality and autonomous driving, to name a few. The completion of these computer jobs must adhere to stringent real-time constraints, and it takes a significant amount of computing resources to bring these tasks to a successful conclusion. Since the volume, weight, and other limitations that restrict vehicles prevent them from being outfitted with powerful computing devices, the computing resources of the onboard devices that are now in use are often not enough to fulfill the processing requirements of these jobs. Install edge servers in the immediate area of the vehicle. Edge computing, in contrast to cloud computing, can provide consumers with computer services that are located relatively near them. Instead of being sent to the cloud, the computing duties that are created by the vehicle are immediately offloaded to the edge server. This reduces the amount of time it takes for computing activities to be transmitted.
As a result, the implementation of edge computing in IOVs is a potential solution to the problem of inadequate processing power shown by vehicles and a means of satisfying the criteria of low latency imposed by tasks.
The offloading of computational duties, in general, may effectively lower the amount of energy that the vehicle requires to operate. Offloading chores is something that consumers are often more likely to do in the interest of keeping the vehicle's battery alive for as long as possible. The number of responsibilities that must be offloaded and carried out inside the Internet of Vehicles will continue to grow because of this. When a significant number of tasks are offloaded and performed, the server is unable to provide computer resources for all the tasks at the same time.
This means that tasks that are not allocated to computing resources must wait to be executed.
At present, it is not possible to disregard the waiting time if the computing jobs that are now queued up to be done have delay requirements. Therefore, to effectively offer computing services for a greater number of offloading jobs, it is important to establish an acceptable scheduling strategy according to the execution time and delay needs of computing tasks.
This paper integrates software-defined networking (SDN) into the Internet of Vehicles, constructs an SDN-assisted computing task offloading system for the Internet of Vehicles in an edge computing environment, and presents a task of computing offloading for vehicles. This is done since SDN can manage network resources more conveniently and effectively. Scheduling model. After that, an improved pointer network is trained using deep reinforcement learning to solve the offload scheduling problem of delay-constrained computing tasks in multi-edge servers on the Internet of Vehicles. This is done in consideration of the complexity of task scheduling and the allocation of computing resources.