.png){kind=icon}
Return
Task offloading in vehicular edge computing based on traffic density-driven task generation
.png){kind=icon}
PRE
AI
DOI：10.1016/j.adhoc.2026.104203.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Existing task offloading methods in vehicular edge computing, though considering the mobility of servers and wireless devices (WDs), often suffer from oversimplified task generation patterns (e.g., static scenarios and fixed device counts) that poorly align with complex real-world traffic environments. Furthermore, these methods exhibit limited adaptability to dynamic network environments. To address these limitations, we propose a vehicular edge task offloading scheme driven by traffic density-generated tasks, in which a congestion coefficient is defined as a comprehensive metric for node load assessment based on two key indicators, road occupancy time and average vehicle speed. First, a task generation mechanism is designed that refines vehicle flow data using a Gaussian distribution to achieve a finer granularity in task generation. To further improve environmental fidelity, a secondary adjustment of task generation is conducted by increasing tasks for nodes with elevating congestion coefficient. Next, a vehicular edge computing system model is constructed, encompassing task generation, transmission, computation, and energy consumption modules. The task offloading problem is then modeled as a Markov Decision Process (MDP), and an optimization strategy based on Deep Deterministic Policy Gradient (DDPG) is developed. In which the defined congestion coefficient is integrated as one of the critical state-space features and a reward function targeting latency-energy consumption co-optimization is constructed. Finally, Simulation results demonstrate that our offloading scheme outperforms existing advanced methods in reducing average Quality of Experience (QoE) while achieving near-optimal solutions.
Keywords:
task offloading
vehicular edge computing
traffic density
congestion coefficient
deep deterministic policy gradient
Journal
IF:
4.8
Papers:
499
Citations:
6.2K
Organization
No organization information available
Cited Papers
ShareSaveShareSaveShareSaveShareSave
ShareSaveAction decoupled SAC reinforcement learning with discrete-continuous hybrid action spaces.png)
NEUROCOMPUTING
NEUROCOMPUTINGIF6.5
PREAI
ShareSaveShareSave
View more