Deep Reinforcement Learning Based User Perception Optimization in 5G Edge Computing Network

China Mobile operates the world’s largest 5G network, with the most significant user base globally. With the application of 5G network technology and Internet of Things (IoT) scenarios becomes more widespread, a large number of devices will connect to China Mobile’s 5G network for communication, interaction and data processing. In computation-intensive scenarios, China Mobile applies edge computing as a supplemental method, offloading user tasks to edge servers to alleviate issues like high latency and bandwidth limitations existent in cloud computing. However, continuous task flows generated by users pose considerable challenges to the limited computation resource of edge servers. In this paper, we investigate dynamic task scheduling oriented towards user perception in 5G edge computing networks. With constraints on computational resources, we have converted the original problem into a user perception quality maximization problem based on the Markov Decision Process (MDP). Considering the uncertainty and dynamics of the state space, we propose a deep reinforcement learning (DRL) based user perception optimization (DUPO) algorithm, introducing an autonomous intelligent decision-making strategy between task computation and cache matching to handle a continuous influx of user tasks. Extensive simulation results indicate DUPO can dynamically adjust the task execution strategy according to random task demand and the system environment, effectively enhancing user perception.