Precaching vehicle selection based on soft actor-critic in CIoV

The rapid development of the Internet of Vehicles and the increasing demand for mobile content services have significantly increased network traffic, leading to congestion and delays. To address these challenges, Content-centric IoV has emerged by integrating IoV with Content-Centric Networking, enabling efficient mobile content delivery. However, CIoV still faces limitations in outage zones where roadside unit coverage is restricted, hindering content transmission. To overcome this issue, we propose a SAC-based Precaching Vehicle Selection scheme that dynamically selects optimal precaching vehicles and determines appropriate content sizes to facilitate content delivery in outage zones. SAC-PVS operates on a snapshot-based inference model, which is specifically designed to optimize precaching decisions at the exact moment of a request without relying on continuous time-series monitoring. It leverages the SAC algorithm to handle continuous action spaces, enabling precise determination of both the optimal caching vehicles and the corresponding content quantities. A hierarchical reward structure penalizes excessive caching and traffic waste while rewarding successful content delivery in outage zones. Simulation results demonstrate that SAC-PVS outperforms both a comparable machine learning approach and a non-machine-learning baseline by reducing content delivery latency and minimizing traffic waste under dynamic vehicular conditions. The proposed SAC-PVS scheme improves the Quality of Service for vehicle users and optimizes network resource utilization for content delivery, offering a scalable and efficient solution for next-generation CIoV content services.