Deep Reinforcement Learning and Markov Decision Problem for Task Offloading in Mobile Edge Computing

doi:10.1007/s10723-023-09708-4

2023-12-04

PRE

Xiaohu Gao *

Mei Choo Ang

Sara A. Althubiti

DOI：10.1007/s10723-023-09708-4

原文链接

原文求助

摘要

En 中文

Mobile Edge Computing (MEC) offers cloud-like capabilities to mobile users, making it an up-and-coming method for advancing the Internet of Things (IoT). However, current approaches are limited by various factors such as network latency, bandwidth, energy consumption, task characteristics, and edge server overload. To address these limitations, this research propose a novel approach that integrates Deep Reinforcement Learning (DRL) with Deep Deterministic Policy Gradient (DDPG) and Markov Decision Problem for task offloading in MEC. Among DRL algorithms, the ITODDPG algorithm based on the DDPG algorithm and MDP is a popular choice for task offloading in MEC. Firstly, the ITODDPG algorithm formulates the task offloading problem in MEC as an MDP, which enables the agent to learn a policy that maximizes the expected cumulative reward. Secondly, ITODDPG employs a deep neural network to approximate the Q-function, which maps the state-action pairs to their expected cumulative rewards. Finally, the experimental results demonstrate that the ITODDPG algorithm outperforms the baseline algorithms regarding average compensation and convergence speed. In addition to its superior performance, our proposed approach can learn complex non-linear policies using DNN and an information-theoretic objective function to improve the performance of task offloading in MEC. Compared to traditional methods, our approach delivers improved performance, making it highly effective for developing IoT environments. Experimental trials were carried out, and the results indicate that the suggested approach can enhance performance compared to the other three baseline methods. It is highly scalable, capable of handling large and complex environments, and suitable for deployment in real-world scenarios, ensuring its widespread applicability to a diverse range of task offloading and MEC applications.

Keyword:

Deep Reinforcement Learning

Deep Deterministic Policy Gradient

Mobile Edge Computing

Task offloading

Markov decision problem