Machine Learning Based Dynamic Pico Base Station On/off Algorithm For Three-tier Heterogeneous Networks

With the rapid development of mobile devices,people's traffic demand has shown an explosive growth trend.The rise of the Internet of Things has also greatly increased the number of devices.In order to meet people's network needs,countries around the world are conducting extensive research and construction for 5G communications.The development of 5G heterogeneous networks greatly increase the throughput of mobile networks by massively deploying different types of base stations.The overlapping coverage of each base station also makes the network more flexible.However,the massive deployment of base stations has brought huge energy consumption while improving people's network experience.5G green communication has become a research hotspot.The current research directions of 5G green communication are mainly reducing wireless interference,improving resource utilization,and unified management of wireless resources.Heterogeneous networks are generally composed of macro base stations,small base stations,and micro base stations.Among them,the number of micro base stations is huge and the deployment is flexible.This article is based on the research of the micro base station opening and closing algorithm.Because the requirements of mobile communication are constantly changing,the micro base station is dynamically opened and closed according to the demand to achieve the effect of energy saving.This paper mainly includes the following three innovative studies:1.Establish a dynamic power model for multi-layer heterogeneous networks.This model calculates the total power of base stations in the cell based on the real-time location of devices,and can allocate bandwidth to ordinary devices and IoT devices with lower demand to achieve energy saving purpose.Based on the model,the expressions for the optimization problem are sorted out.2.A multi-layer heterogeneous network model is proposed based on a convolutional neural network for pico base station on/off algorithm.This algorithm has low complexity and can calculate the pico base station on/off states in real time.A search algorithm is proposed to obtain the approximate optimal solution of the pico base station's on/off states to generate a training set.The experimental results show that the energy saving effect of the pico base station on/off algorithm based on the convolutional neural network can reach 83.4%of the best energy saving effect.3.Aiming at the multi-layer heterogeneous network model,a micro base station on/off states algorithm based on asynchronous advantage actor-critic algorithm was proposed.This algorithm belongs to the field of reinforcement learning,so there is no need to generate a training set in advance,the training time is short and the effect is excellent.During the implementation of the algorithm,the pico base station on/off state can be obtained within a few seconds.The experimental results show that the energy saving effect of the pico base station on/off algorithm based on the asynchronous advantage actor-critic algorithm can reach 92.1%of the best energy saving effect.