.png){kind=icon}
.png){kind=icon}
Return
A vehicular edge computing offloading and task caching solution based on spatiotemporal prediction
.png){kind=icon}
PRE
AI
DOI：10.1016/j.future.2024.107679.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Traditional research on vehicular edge computing often overlooks the need for large amounts of real-time data with temporal and spatial characteristics. Existing task offloading strategies mainly use a binary approach, neglecting the limited vehicle computing resources and failing to utilize vehicle and edge server resources fully. To this end, this paper proposes a spatio-temporal prediction and Deep Reinforcement Learning (DRL) based task offloading and caching scheme for in-vehicular edge computing. In this paper, a digital twin (DT)-assisted vehicular edge environment is designed to utilize DT technology to obtain a large amount of real-time environmental data for DRL computation. Meanwhile, considering that the data in the in-vehicular edge environment has temporal and spatial characteristics, this paper proposes a demand prediction model based on a spatio-temporal graph neural network (STGNN). This paper proposes a task caching model based on the improved A3C algorithm, which caches tasks based on the prediction results to reduce the computation. In addition, this paper proposes a computational model based on Deep Neural Network (DNN) partitioning and DNN branching, where computational tasks are partitioned and partial offloading is used to realize the full use of computational resources. Since the time scale of task timeliness is much larger than that of vehicle mobility and network state changes, this paper models the problem using a Markov decision process with dual time scales. Experimental results show that this scheme can effectively reduce the delay of task processing.
Keywords:
Edge computing
Internet of Vehicles
Deep reinforcement learning
STGNN
DNN inference task
Digital twin
Calculate offloading
Task caching
Journal
F
IF:
6.1
Papers:
6.8K
Citations:
2.3W