Deep Reinforcement Learning Based Dynamic Reputation Policy in 5G Based Vehicular Communication Networks

Vehicular networks are vulnerable to various attacks from malicious vehicles or infrastructures within a network. The collaborative misbehavior detection system can be used to detect these internal or insider attacks. However, in a collaborative misbehavior detection system, an attacker may lower th...

Full description

Saved in:
Bibliographic Details
Published inIEEE transactions on vehicular technology Vol. 70; no. 6; pp. 6136 - 6146
Main Authors Gyawali, Sohan, Qian, Yi, Hu, Rose
Format Journal Article
LanguageEnglish
Published New York IEEE 01.06.2021
The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
Subjects
5G mobile communication
Collaboration
Communication networks
Deep learning
deep reinforcement learning
Edge computing
Misbehavior detection
Reinforcement learning
Reliability
Reputations
Roads
security
Servers
strong trust
Vehicle dynamics
vehicle reputation
Vehicles
vehicular communication networks
weak trust
Online AccessGet full text

Cover

More Information
Summary:Vehicular networks are vulnerable to various attacks from malicious vehicles or infrastructures within a network. The collaborative misbehavior detection system can be used to detect these internal or insider attacks. However, in a collaborative misbehavior detection system, an attacker may lower the detection accuracy by sending false feedback. A trust model can be used to stimulate vehicles to send true feedbacks. However, an attacker can take advantage of weak or strong reputation update methods. A dynamic trust or reputation update policy can be used to stimulate vehicles to send true feedbacks. In this paper, we propose a deep reinforcement learning based dynamic reputation update policy. In the proposed scheme, feedbacks from vehicles are combined in vehicular edge computing (VEC) servers using Dempster-Shafer theory and the results are used to predict the average number of true messages. VEC then uses deep reinforcement learning to determine the optimum reputation update policy to stimulate vehicles to send true feedbacks. In addition, through extensive simulations, we show that the proposed dynamic reputation-policy is better in terms of the average number of true feedbacks compared to the existing reputation update policy.
ISSN:0018-9545
1939-9359
DOI:10.1109/TVT.2021.3079379