.png){kind=icon}
.png){kind=icon}
Return
Generative diffusion model-based QMIX for joint task offloading and resource allocation in VEC systems
.png){kind=icon}
PRE
AI
DOI：10.1016/j.comnet.2025.111516.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
To cope with the high computing demand and latency requirements of emerging vehicular applications, vehicular edge computing (VEC) has been regarded as a promising computing paradigm that improves vehicular performance by introducing edge computation offloading for resource-constrained vehicles. Compared to the conventional delay metric, information freshness is more crucial for applications, such as automatic driving, auto navigation, etc., which can effectively avoid potential accidents caused by outdated data. Therefore, we apply the age of information (AoI) to measure the freshness of all vehicles’ tasks. Then, a long-term average AoI minimization problem is formulated by jointly optimizing the edge-cloud cooperation task offloading and resource allocation under time-varying environments. To solve this problem, we propose an optimization-oriented multi-agent deep reinforcement learning (MADRL) framework. Specifically, we propose a generative diffusion model (GDM)-based value function decomposition MADRL algorithm, named GDM-QMIX, to learn power allocation and offloading policies for multiple vehicle agents. Meanwhile, the closed-form solution of the wired transmission rate and computing resources allocation is derived based on Karush-Kuhn–Tucker (KKT) conditions to evaluate the quality of actions of GDM-QMIX, thereby avoiding a huge action space and achieving joint optimization. Simulation results demonstrate the effectiveness of the proposed algorithm in solving the dynamic task offloading and resource allocation problem and the superiority of the proposed algorithm over the benchmark schemes in terms of the average AoI.
Keywords:
vehicular edge computing
age of information
task offloading
resource allocation
multi-agent deep reinforcement learning
Journal
IF:
4.6
Papers:
1.5K
Citations:
1.6W