D3QN-based collaborative offloading algorithm for vehicular networks assisted by digital twins
To address the challenges of highly dynamic topologies, task diversity, and low-latency constraints in vehicular edge computing, a D3QN-based collaborative offloading algorithm assisted by digital twin technology was proposed. Firstly, a digital Siamese network was constructed to realize the dynamic...
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | zho |
| Published: |
Editorial Department of Journal on Communications
2025-01-01
|
| Series: | Tongxin xuebao |
| Subjects: | |
| Online Access: | http://www.joconline.com.cn/zh/article/doi/10.11959/j.issn.1000-436x.2025144/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | To address the challenges of highly dynamic topologies, task diversity, and low-latency constraints in vehicular edge computing, a D3QN-based collaborative offloading algorithm assisted by digital twin technology was proposed. Firstly, a digital Siamese network was constructed to realize the dynamic modeling of traffic state, which integrated vehicle spatio-temporal and resource information, was different from the traditional static clustering strategy, improved the stability of collaborative clustering and reduced the strategy search space. Next, based on task decomposability, two types of task models were established, and a hybrid offloading strategy was devised to accurately adapt to dynamic real-world demands. Furthermore, a D3QN-based collaborative offloading algorithm for vehicle clusters was developed, utilizing a dual-network architecture to decouple action and target evaluations, thereby suppressing <italic>Q</italic>-value bias, accelerating policy convergence, and achieving a balance between utility and latency. The simulation results demonstrate that the proposed scheme can significantly reduce task processing latency in high-dynamic and high-load scenarios, and achieves an average system utility improvement of 5.32%, 8.54%, 1.47%, 11.2%, 68.51%, and 103.15% compared to the other six baseline algorithms, respectively. |
|---|---|
| ISSN: | 1000-436X |