A Deep Reinforcement Learning-Based Technique for Enhancing Cache Hit Rate by Adapting to Dynamic File Request Patterns
Mingoo Kwon, Minseok Song · Korean Institute of Smart Media · 2025
Improving the cache hit ratio in edge caching is crucial for effectively handling file requests within limited cache capacity and optimizing network and system resources. This study proposes a file cache management approach based on Deep Reinforcement Learning (DRL). The proposed method aims to enhance data access performance by efficiently utilizing limited cache resources, adapting to dynamic file request patterns, and improving the cache hit ratio. Specifically, it employs the Proximal Policy Optimization (PPO) algorithm to intelligently manage cache replacement policies and effectively handle the complexity of continuous action spaces. The PPO network consists of an action space for caching decisions, an observation space reflecting changes in file access patterns, and a reward model evaluating the costs and benefits of various caching strategies. Experimental results demonstrate that the proposed method improves the cache hit ratio by an average of 2.28% and up to 16.25% compared to traditional greedy algorithms, validating the effectiveness of the DRL-based approach in constrained cache environments.