Reinforcement Learning Minimizes Payment Costs in Internet of Vehicles Caching

With the rapid advancement of Internet of Vehicles technology, various in-vehicle applications requiring extensive data transmission and computational tasks have emerged. The vehicular edge caching system facilitates efficient content delivery for in-vehicle applications by storing content at roadside units, alleviating the load on backhaul links. However, the limited storage space of roadside units underscores the critical importance of formulating appropriate caching strategies. Considering the actual characteristics of the Internet of Vehicles scenario, we delve into the issue of user-centric content delivery. The problem of limited roadside unit collaborative caching replacement is modeled as a Markov decision process (MDP). A novel edge caching replacement algorithm (CR-DQN) based on Deep QNetwork is proposed to dynamically allocate and update the cached content of RSU in real time while ensuring service quality. The vast simulation results demonstrate that the algorithm proposed exhibits significant performance improvements in hit rate and payment costs compared to other baseline schemes.