Adaptive Edge Caching in Dynamic Environments Using PPO and Transfer Learning

This study tackles the problem of edge caching within dynamic settings, where increasing traffic demands put pressure on backhaul links and core network infrastructures. We introduce a caching method based on Proximal Policy Optimization (PPO) that integrates essential file characteristics, including size, lifetime, importance, and popularity, while also accommodating random file request patterns to better mirror real-world edge caching situations. Dynamic environments often experience fluctuations in content popularity and request rates, rendering previously established policies less effective since they were tailored to earlier conditions. Although training a new policy from scratch in a changed environment is feasible, it is often inefficient and resource-intensive. To solve this issue, we present a PPO algorithm enhanced with transfer learning, which improves convergence in new environments by utilizing previously acquired knowledge. Our simulation results highlight the substantial advantages of our approach, outperforming recent transfer learning-based methods in terms of convergence rate, demonstrating its potential to enhance edge caching in dynamic real-world scenarios.