.png){kind=icon}
.png){kind=icon}
Return
Clustering-Driven Cooperative Caching and Trajectory Optimization Based on Dual-Agent Reinforcement Learning in UAV-Assisted VANETs
.png){kind=icon}
PRE
AI
DOI：10.1109/tccn.2026.3705803.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
The high mobility and heterogeneity of Vehicular Networks (VANETs) impose significant challenges for edge caching and resource optimization, as rapidly changing topologies and diverse content demands hinder real-time responsiveness. To address these issues, we propose an integrated optimization framework combining dynamic clustering, popularity prediction, and joint caching-trajectory control. Firstly, we design a location-interest-based dynamic vehicle clustering mechanism that groups vehicles according to communication reachability and semantic content similarity, forming stable and service-oriented clusters. Secondly, a hierarchical asynchronous federated prediction model is developed to aggregate local cluster pReferences with global request trends, achieving timely, accurate, and privacy-aware popularity estimation. Finally, we construct a Dual-Agent Collaborative Caching and Trajectory Optimization (DACTCO) framework, where a Dueling Double Deep Q-Network (D3QN) agent governs UAV-assisted caching decisions while a Deep Deterministic Policy Gradient (DDPG) agent adapts UAV positions. Through a unified scalarized reward, the two agents jointly minimize average delay and energy consumption in a coordinated manner. Simulation results demonstrate that the proposed framework significantly enhances service quality and resource efficiency, achieving reductions of up to 6.3% in average delay, 42% in average energy consumption, and improvements of up to 36.2% in cache hit rate compared with state-of-the-art baselines.
Keywords:
Cooperative edge caching
dynamic vehicle clustering
hierarchical asynchronous federated learning
dual-agent reinforcement learning
UAV trajectory optimization
Journal
I
IF:
7
Papers:
1.5K
Citations:
5.5K