Research on Optimal Computing Offloading Strategy of Internet of Vehicles Based on Reinforcement Learning.

Saved in:

Bibliographic Details
Title:	Research on Optimal Computing Offloading Strategy of Internet of Vehicles Based on Reinforcement Learning.
Authors:	Qiu, Shaolin¹ bykjsl@163.com, Deng, Shuchao² dengsc@ahut.edu.cn, Pang, Honglei³ panghl@niit.edu.cn, Wang, Bo⁴ 3181487830@qq.com, Ye, Hao⁵ 18913929898@163.com
Source:	IAENG International Journal of Computer Science. May2026, Vol. 53 Issue 5, p2011-2020. 10p.
Subjects:	Edge computing, Federated learning, Multiagent systems, Privacy, Mobile communication systems, Reinforcement learning
Abstract:	Vehicle edge computing (VEC) effectively alleviates the computational resource limitations of in-vehicle terminals by offloading computation-intensive tasks to edge servers. However, offloading decision-making remains challenging due to highly dynamic network topologies, complex task dependency structures, and multi-dimensional quality-of-service (QoS) constraints. To address these issues, this paper proposes a Federated Multi-Agent Proximal Policy Optimization (Fed-MAPPO) framework for intelligent computation offloading with multi-objective trade-offs. The proposed approach models task dependencies using directed acyclic graphs (DAGs) and formulates the problem as a partially observable Markov decision process (POMDP). A dependency-aware state encoder is designed to jointly capture task topology and network dynamics, while a differential privacy-based secure aggregation mechanism is introduced to protect sensitive vehicular data during federated learning. In addition, a hierarchical attention-based value decomposition network is employed to resolve the credit assignment problem in multi-vehicle collaboration. Experiments conducted on a SUMO/TraCI and OMNeT++ co-simulation platform with realistic traffic scenarios generated from Shanghai taxi trajectories demonstrate that, compared with FedAvg-DQN, Fed-MAPPO improves the task completion rate by 12.7%, reduces average latency by 23.4%, accelerates convergence speed by 35.6%, and maintains performance degradation below 3% under strong privacy constraints (ε = 0.5). These results indicate that the proposed framework provides a scalable and robust paradigm for intelligent offloading in large-scale heterogeneous vehicular networks. [ABSTRACT FROM AUTHOR]
	Copyright of IAENG International Journal of Computer Science is the property of International Association of Engineers (IAENG) and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)
Database:	Engineering Source

Description
Abstract:	Vehicle edge computing (VEC) effectively alleviates the computational resource limitations of in-vehicle terminals by offloading computation-intensive tasks to edge servers. However, offloading decision-making remains challenging due to highly dynamic network topologies, complex task dependency structures, and multi-dimensional quality-of-service (QoS) constraints. To address these issues, this paper proposes a Federated Multi-Agent Proximal Policy Optimization (Fed-MAPPO) framework for intelligent computation offloading with multi-objective trade-offs. The proposed approach models task dependencies using directed acyclic graphs (DAGs) and formulates the problem as a partially observable Markov decision process (POMDP). A dependency-aware state encoder is designed to jointly capture task topology and network dynamics, while a differential privacy-based secure aggregation mechanism is introduced to protect sensitive vehicular data during federated learning. In addition, a hierarchical attention-based value decomposition network is employed to resolve the credit assignment problem in multi-vehicle collaboration. Experiments conducted on a SUMO/TraCI and OMNeT++ co-simulation platform with realistic traffic scenarios generated from Shanghai taxi trajectories demonstrate that, compared with FedAvg-DQN, Fed-MAPPO improves the task completion rate by 12.7%, reduces average latency by 23.4%, accelerates convergence speed by 35.6%, and maintains performance degradation below 3% under strong privacy constraints (ε = 0.5). These results indicate that the proposed framework provides a scalable and robust paradigm for intelligent offloading in large-scale heterogeneous vehicular networks. [ABSTRACT FROM AUTHOR]
ISSN:	1819656X