Task offloading and proactive resource allocation in vehicular edge computing via reinforcement learning

Karimi, Elham (2022) Task offloading and proactive resource allocation in vehicular edge computing via reinforcement learning. Doctoral (PhD) thesis, Memorial University of Newfoundland.

[img] [English] PDF - Accepted Version
Available under License - The author retains copyright ownership and moral rights in this thesis. Neither the thesis nor substantial extracts from it may be printed or otherwise reproduced without the author's permission.

Download (7MB)


Given the rapid increase of various applications in vehicular networks, it is crucial to consider a exible architecture to improve the Quality-of-Service (QoS). Utilizing Multi-access Edge Computing (MEC) as a distributed paradigm with computation capabilities closer to the vehicles can be a promising solution to reduce response time in such a network. However, MEC nodes are deprived of processing all tasks offloaded by the vehicles and suffer from limited resources compared with the central cloud. The offloaded tasks usually have different priorities for processing and various resource demands, and they even are dropped when the response time for task processing expires. Due to the workload dynamics at MEC nodes and the randomness of task arrivals, it is challenging to determine proper MEC servers for task offloading and how to manage resources for the application's demands. This dissertation proposes cooperation between MEC and central cloud decisions for different vehicular application offloading. First, we formulate a new NP-hard resource allocation problem to guarantee the required response time. We transform the environment into a finite Markov decision process that only depends on the current state and action space. We utilize deep reinforcement learning, a proper computational model, to automatically learn the dynamics of the network state and rapidly capture an optimal solution based on the current state and action space. In this dissertation, we develop an intelligent workload prediction of each MEC node utilizing Multivariate Long Short-Term Memory (LSTM) to propose a proactive resource allocation algorithm for various tasks in a dynamic vehicular network. In our algorithm, we classify tasks based on their priorities and migrate the tasks with lower priority to provide service for those with higher priority. Moreover, we apply distributed deep reinforcement learning to solve our problem to increase the efficiency and accuracy of the proactive resource allocation algorithm. Extensive numerical analysis and results illustrate how our proposed algorithms can increase the ratio of accepted high-priority tasks and reduce response time.

Item Type: Thesis (Doctoral (PhD))
URI: http://research.library.mun.ca/id/eprint/15901
Item ID: 15901
Additional Information: Includes bibliographical references (pages 97-105)
Keywords: vehicular edge computing, proactive resource allocation, task offloading, deep reinforcement learning, workload prediction, multivariate LSTM
Department(s): Science, Faculty of > Computer Science
Date: August 2022
Date Type: Submission
Digital Object Identifier (DOI): https://doi.org/10.48336/RFRB-4P06
Library of Congress Subject Heading: Edge computing; Reinforcement learning; Resource allocation; Computer networks; Deep learning (Machine learning)

Actions (login required)

View Item View Item


Downloads per month over the past year

View more statistics