Page 107 - ITU Journal Future and evolving technologies Volume 2 (2021), Issue 4 – AI and machine learning solutions in 5G and future networks

ITU Journal on Future and Evolving Technologies, Volume 2 (2021), Issue 4

APPLYING MACHINE LEARNING IN NETWORK TOPOLOGY OPTIMIZATION
Zhouwei Gang, Qianyin Rao, LinGuo, LinXi, Zezhong Feng, QianDeng

GuanShanXi Road, Guanshan Lake District, GuiYang GuiZhou, China

NOTE: Corresponding author: Zhouwei Gang (13595101082@139.com)

Abstract – Nowadays, telecommunications have become an indispensable part of our life, 5G technology
brings better network speeds, helps the AR and VR industry, and connects everything. It will deeply change
our society. Transmission is the vessel of telecommunications. While the vessel is not so healthy, some of
them are overloaded, meanwhile, others still have lots of capacity. It not only affects the customer experience,
but also affects the development of communication services because of a resources problem. A transmission
network is composed of transmission nodes and links. So that the possible topology numbers equal to node
number multiplied by number of links means it is impossible for humans to optimize. We use Al instead of
humans for topology optimization. The AI optimization solution uses an ITU Machine Learning (ML)
standard, Breadth-First Search (BFS) greedy algorithm and other mainstream algorithms to solve the
problem. It saves a lot of money and human resources, and also hugely improves traffic absorption capacity.
The author comes from the team named"No Boundaries". The team attend ITU AI/ML in 5G Challenge and
won the Gold champions (1st place).

Keywords – 5G, artificial intelligence (Al), data handling, intelligence level,machine learning (ML)

1. BACKGROUND Mobile Kaili PTN network as an example. Among the
913 links in the entire network, high-load links
Nowadays, telecommunications have become an account for 9% while low-load links reach 53%.
indispensable part of people’s lives. We connect to Operations staff need to optimize the network
the Internet through mobile phones for social topology.
networking (Facebook, Weibo, etc.), entertainment
(YouTube, mobile games, etc.), e-shopping (Amazon,
eBay, etc.) and work (remote office OA, etc.), and the
upcoming 5G network has expanded from people-
to-people communication to people-to-things, and
things-to-things communication. Therefore, the
future network will be more deeply embedded in
our society and life. Assuming that a tele-
communication system is a human body,
transmission is as important as a vessel system
which sends our data (also known as traffic) to the
place it is needed. To construct a highly efficient Fig. 1 – Link utilization
network, each node should be used efficiently to A transmission network is composed of
form the transmission network. transmission nodes and links. So that the possible
However, just as a large number of diseases are topology numbers equal the node number
related to a blood vessel ， the situation of the multiplied by the number of links, taking Kaili as an
transmission network is not optimistic in the actual example, the possibility of all topologies exceeds 2.9
communication network. million, which is impossible for humans to optimize.

Due to the characteristics of mobile So, at present, we can only adopt two methods, one
communications, increasingly complex mobile is to adjust only one link locally. We optimize some
Internet services, and increasing traffic, the of the chains with ultra-high utilization rates by
utilization rate of the transmission network is experience or artificial prediction to reduce the
generally unbalanced. Congestion is caused by the number of nodes and load on a single link. And the
overload of a small number of key nodes, and about other is to directly upgrade the hardware, which
half of the nodes are under low load. Take the China increases the bandwidth capacity of the entire link

© International Telecommunication Union, 2021 91