Joint Computation Offloading and Resource Allocation for Edge-Cloud Collaboration in Internet of Vehicles via Deep Reinforcement Learning

• Published in: IEEE systems journal Vol. 17; no. 2; pp. 2500 - 2511
• Main Authors: Huang, Jiwei, Wan, Jiangyuan, Lv, Bofeng, Ye, Qiang, Chen, Ying
• Format: Journal Article
• Language: English
• Published: New York IEEE 01.06.2023; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Algorithms; Bandwidth; Cloud computing; Computation offloading; Computational modeling; Deep learning; deep reinforcement learning (DRL); Delay; Delays; Edge computing; Heuristic algorithms; Internet of Vehicles; Internet of Vehicles (IoV); Markov processes; Mobile computing; mobile edge computing; Optimization; Resource allocation; Roads; Task analysis; Vehicle dynamics
• ISSN: 1932-8184; 1937-9234
• DOI: 10.1109/JSYST.2023.3249217

Mobile edge computing (MEC) and cloud computing (CC) have been considered as the key technologies to improve the task processing efficiency for Internet of Vehicles (IoV). In this article, we consider a random traffic flow and dynamic network environment scenario where MEC and CC are collaborated for processing delay-sensitive and computation-intensive tasks in IoV. We study the joint optimization of computation offloading and resource allocation (CORA) with the objective of minimizing the system cost of processing tasks subject to the processing delay and transmission rate constraints. To attack the challenges brought by the dynamic environment, we use the Markov decision process model for formulating the dynamic optimization problem, and apply a deep reinforcement learning (DRL) technique to deal with high-dimensional and continuous states and action spaces. Then, we design a CORA algorithm, which is able to effectively learn the optimal scheme by adapting to the network dynamics. Extensive simulation experiments are conducted, in which we compare the CORA algorithm with both non-DRL algorithms and DRL algorithms. The experimental results show that the CORA algorithm outperforms others with excellent training convergence and performance in processing delay and processing cost.