摘 要

随着科技的飞速发展，通信技术也是日新月异。从最初的第一代蜂窝网络，再到后来的2G网络、3G网络，直至如今普及开来的4G网络，以及未来的5G网络，用户的通信体验也越来越好，上网质量越来越高。为了进一步降低用户的上网延迟时间，提升服务质量，在现有基站的基础上引入了小基站，从而形成了异构网络。异构网络虽然明显的提升了通信质量与上网质量，但数量众多的小基站所带来的能量消耗不可忽视。节能减排已经成为了各行各业的共识，通信行业耗能巨大，并且主要是来自于移动蜂窝网络的耗能，其中基站的耗能又占了一半以上。因此，降低异构网络的能耗意义重大且势在必行。

本课题主要针对目前异构密集网络带来的大量能量消耗问题，研究超密集小区网络部署情形下的绿色网络的系统能效优化方案即研究绿色网络技术。文章选取了通过人工智能技术来实现网络绿色化，提升系统能效。首先模拟异构超密集小区场景构建数据集，设定小区内各个微微基站在二维平面内按照独立泊松点过程分布，并在小区覆盖范围内随机地撒入用户。通过遍历异构小区内每一个基站关闭时的小区吞吐量与全开时的小区吞吐量进行对比，当关闭该基站性能更好时输出此时的小区吞吐量、小区内用户总干扰功率和小区用户总接受功率，反之则输出该基站开，通过多次模拟该场景得到大量带有基站开关标签和小区吞吐量，总接受功率，干扰功率的数据。然后将这些训练数据带入预设的神经网络中训练网络，再通过调整参数使结果更加准确，得到一个可以预测特定情况下能达到使系统能效更高目的的各基站开关情况的神经网络。大量仿真结果显示，在此基站休眠方案下，系统耗能比不使用任何节能方案的网络明显降低，同时保证了用户服务质量（Qos）。本课题提出的系统能效优化方案能够有效地应用于超高密度小区网络的设计中。

关键词：5G 绿色网络；基站休眠；系统能效；神经网络

Abstract

Communication technology is getting better and better with the rapid development of technology. From the initial first-generation cellular network to the later 2G network, 3G network, to the popular 4G network, and the future 5G network, the user's communication experience is getting better and better, and the quality of the Internet is getting higher and higher. In order to further reduce the user's Internet access delay time and improve the quality of service, a small base station is introduced on the basis of the existing base station, thereby forming a heterogeneous network. Although the heterogeneous network obviously improves the communication quality and the quality of the Internet, the energy consumption brought by a large number of small base stations cannot be ignored. Energy-saving and emission reduction has become the consensus of all walks of life. The communication industry consumes a lot of energy, and mainly comes from the energy consumption of mobile cellular networks, in which the base station's energy consumption accounts for more than half. Therefore, reducing the energy consumption of heterogeneous networks is significant and imperative.

In this paper, we focus on the green network energy efficiency in high-dense small cell networks to solve the energy consumption problem that the heterogeneous ultra-dense network brings. We optimize the system energy efficiency by the base station sleep strategy, and A new mathematical model is mentioned in this paper. Firstly, we will derive a random heterogeneous ultra-dense cell network model, assuming that each base station is distributed based on a separate two-dimensional homogeneous Poisson point process. Then, we put the user equipment among cells randomly. Secondly, we compare the Total cell throughput when a base station is under the sleeping mode with the total cell throughput when all stations in the heterogeneous cell network are under active mode, and we get the total cell throughput, the received power of user equipment and the interference power of user equipment when the former is bigger, otherwise we consider that the target station is under active mode. Then a lot of data with base station switch label, total cell throughput, total received power and interference power can be obtained by simulating the scene for many times. We can put these training data into the neural network and adjust the parameters to make the results more accurate, so that a neural network is obtained to predict the switching conditions of each base station which can achieve the goal of optimizing the system energy efficient under specific circumstances. A large number of simulation results show that under this base station sleep scheme, the system energy consumption is significantly lower than that of the network without using any energy-saving scheme, and the user service quality (Qos) is guaranteed. So our optimal sleeping strategy can be applied to design the small cell network.

Key Words: 5G; Green network Base station sleep; System energy efficiency; Neural Networks

目 录

第1章 绪论 1

1.1 课题的研究背景和意义 1

1.2 国内外绿色网络发展现状 3

1.3 论文结构安排 4

第2章 绿色网络的评价及节能方案的确定 5

2.1 绿色网络节能策略与节能方案的选取 5

2.1.1 移动通信网络的能耗分析 5

2.1.2 绿色节能策略 5

2.1.3 节能方案的选取 7

2.2 基站休眠方案 7

2.2.1 基站休眠策略概况 7

2.2.2 基站耗能分析 8

2.2.3 休眠策略分析 9

2.2.4 本课题的基站休眠方案 10

2.3 本章小结 10

第3章 基于仿真平台创建数据集 11

3.1 仿真试验平台介绍 11

3.1.1 仿真平台的模拟场景 11

3.1.2 仿真平台的模块 12

3.2 创建数据集 14

3.2.1 实验设置 14

3.2.2 以平均距离和用户数作为输入 15

3.2.3 以吞吐量和接收功率干扰功率作为输入 16

3.3 本章小结 18

第4章 基于神经网络的基站休眠策略设计与实现 19

4.1 BP神经网络概述 19

4.1.1 神经网络介绍 19

4.1.2 BP神经网络的结构 20

4.2 基于BP神经网络的基站休眠模型的设计 22

4.2.1 数据预处理 22

4.2.2 BP神经网络模型的结构设计 23

4.2.3 BP神经网络模型的训练 23

4.3 基于BP神经网络的预测基站休眠模型的实现 24

4.3.1 BP神经网络的仿真结果 24

4.3.2 GUI界面的设计 25

4.4 本章小结 25

第5章 基于BP网络模型基站休眠策略的网络能效仿真评估 26

5.1 基站休眠节能算法设计思路 26

5.2 网络能效系统模型 27

5.3 网络能效的仿真结果 29

5.4 本章小结 31

第6章 结束语 32

6.1 全文工作的总结 32

6.2 未来研究工作的展望 32

参考文献 33

致 谢 35

第1章 绪论

1.1 课题的研究背景和意义

随着科学技术的发展以及人们生活质量的提升，网络的智能性已成为当代社会创新的重要指标，人们对网络的需求也在不断增加。具体表现在从原始的读书看报到现在的手机网页浏览新闻，从原来的现金纸币交易到现在的手机收付款码，处处离不开以手机为载体的网络交互。传统的通过计算机上网，从被发明起到现在已经经历了四代历史性革新，现在便携式笔记本电脑在办公、教学场所随处可见，新兴的通过手机上网热潮势不可挡，现在已经成为了绝大多数人每天必备的上网工具，可以预见将来通过手机进行网络交互会普及到百分百涵盖。每一代移动技术的发展时间在10年左右，不断的创新推动着网络导向下一个更优的平台。

移动通信有着快速的发展史，如图1.1所示。