Research On Iot Transient Data Caching Strategy Based On Deep Reinforcement Learning

With the development of the Internet of Things(IoT),the rapid growth of IoT devices and traffic has brought tremendous pressure to the network.In order to meet the challenges of IoT,edge caching,by caching popular data at edge nodes,could effectively offload network traffic,reduce network latency,and improve quality of experience.However,edge caching in IoT scenarios has two major features: First,there are many transient data with time-sensitive requests in IoT.Second,cache space of the edge nodes is limited.Based on the features of IoT caching,this paper studies the transient data caching strategy and proposes an IoT transient data caching strategy based on deep reinforcement learning(DRL).Firstly,the data of time-sensitive requests in the IoT is defined as transient data,and the transient data is modeled.The main properties and characteristics of transient data,such as lifetime,freshness and freshness loss,are analyzed.For the edge cache scenarios,the cost of acquiring transient data is composed of the communication cost and freshness loss cost.A transient data storage structure suitable for edge cache nodes is designed.Secondly,the process of processing the IoT transient data requests by the edge node is analyzed.The caching replacement strategy is formulated as a Markov process,and the DRL algorithm is used to solve the cache replacement problem.A suitable DRL algorithm is designed for the caching scenario to capture popularity change and time features,optimize the communication cost and freshness loss cost of obtaining IoT transient data,and learn a flexible and efficient caching strategy.Finally,the proposed caching algorithm and system based on DRL are implemented,and the proposed transient data caching strategy is tested and compared performance with two baseline algorithms in simulation environment.By modeling and analyzing the edge caching strategy of IoT transient data,a caching replacement algorithm based on deep reinforcement learning is proposed.The simulation results show that the proposed caching strategy,compared with traditional caching strategies,can learn more flexibly and efficient caching strategies and optimize the overall efficiency of transient data caching in different scenarios.