Page 108 - 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

by expanding hardware boards and ports to reduce Although no mature solution was found, the team
link utilization (traffic/bandwidth). Due to the was inspired to use machine learning instead of
shortcomings and limitations of the current scheme, humans for topology optimization.
it is difficult to improve the current unbalanced
transmission network. 2. SOLUTION

2.1 Simulated environment
In order to explore the optimization of network
topology based on machine learning, China Mobile
selected more than 2,000 pieces of node topology
information in three cities A, B, and C from the real
network and the traffic situation of the nodes in the
previous 20 days as the basic data of the contest,
and the comprehensive evaluation indexes of
topology optimization results (single day) are as
follows:
maxobject_ratio avg − (E avg + E min + E max ) − α ∙ sub_ratio − β ∙ hang_ratio
Fig. 2 – Solution1:Local(Ring) Adjustment: "big-to-small"
optimization for some links, reducing the number of nodes object_ratio avg : the hourly daily average value of the
and load ratio of link bandwidth utilization optimized to the
target range;
E avg : hourly daily average value of the network
topology load balancing E value;

E min : hourly daily minimum value of the network
topology load balancing E value;

E max : hourly and daily maximum value of the
network topology load balancing E value;

sub_ratio : ratio of the number of nodes on the
secondary link in each link to the number of nodes
in the entire link (hourly daily average);

hang_ratio : the ratio of the number of nodes
Fig. 3 – Solution2:Node expansion: Increase bandwidth of the hanging in each link to the number of nodes in the
link by upgrading hardware, and reduce link utilization
entire link (hourly daily average);
To solve this problem more effectively, the No
Boundaries team has read a lot of literature, but α，β:Constant coefficients (α=0.02, β =0.05 ).
there is no mature solution. The specific algorithm of the evaluation index is as
Qianyin Rao believes that applying the new SDN follows:
technology of 5G transmission network to Target range of link bandwidth utilization:
transmission network can only solve the problems
3
of delay and local optimization, and cannot support [ × ， 7 × ]
global topology optimization[1]. Intelligent control 5 5
is still insufficient.
The calculation method of the load balancing E
Lin Guo believes that wireless devices are the edges of value is as follows:
networks, which differ from the nature of the E value = variance of the fp value of all links in the
transmission equipment[2] [3]. So a wireless network network topology
topology optimization scheme cannot be adopted.
∑ nodes
Zhouwei Gang believes that the number of nodes in Link bandwidth utilization =

the data center and Internet scenarios is far less
than that of the transmission network and cannot
be applied[4] [5] [6].

92 © International Telecommunication Union, 2021