Deep Reinforcement Learning Based Real-Time Proactive Edge Caching in Intelligent Transportation System

Edge content cache (ECC) is considered a key technology in Intelligent Transportation Systems (ITS). It can meet the growing demand for computing-intensive and latency-sensitive vehicular applications while improving communication efficiency. Existing ECC schemes can only achieve cache prefetch by predicting content demands in the future time period, or merely implement cache replacement based on current requests, however, ignoring the future. These schemes still have the potential to improve performance with both current and future request context information combined. We propose a real-time proactive edge content caching using deep reinforcement learning. This new scheme trains cache servers to perform cache prefetch in real time on each request, thus combining the cache prefetch and cache replacement without requiring users’ privacy information. In addition, it enables collaborations between multiple location-related cache servers. This new solution can improve the efficiency of ECC and reduce the overall communication cost while protecting user privacy. Simulation results indicate that this scheme outperforms other baseline schemes in communication cost and cache hit ratio.