Deep reinforcement learning based big data resource management for 5G/6G communications

With the advent of the Internet of Everything era, communication data has exploded, which requires more communication resources, such as frequency, time, and energy. In this context, this paper presents a machine learning-based data packet scheduling scheme to achieve efficient data packet transmission in the 5G/6G communication systems. To minimize the average number of packet overflows (APNO), we propose distributed deep deterministic policy gradient (DDPG)-based algorithm for multidimensional resource scheduling. To improve the algorithm stability and training efficiency, the strategy of centralized training and distributed execution is adopted, and an Action Adjuster is designed. The proposed algorithm enables the multidimensional resource management of the 5G/6G commu-nication systems without any information interaction between each agent. Simulation results show that the proposed Action Adjuster DDPG algorithm achieves faster convergence and less data overflow compared to other benchmark algorithms.