A Deep Reinforcement Learning-Based Caching Strategy for Internet of Things

With the continuous growth of the Internet of Things (IoT), the specific needs of these networks are becoming more evident. Transient data generated and limited energy resources are two of the characteristics of IoT networks that impose some limitations. Moreover, the conventional quality of service requirements, such as minimum delay, are still needed in these networks. By implementing an effective caching policy, it is possible to meet the current demands while easing the specific limitations of IoT networks. By leveraging deep reinforcement learning technique, without the need of prior knowledge of the contents' popularity, contents lifetimes or any other type of contextual information, we have managed to develop a caching policy which increases the cache hit rate and decreases the energy consumption of IoT devices while simultaneously considering the limited lifetime of the data contents. The simulation results show that our proposed method outperforms the conventional Least Recently Used (LRU) method by considerable margins in all aspects.